test(coverage): add Behave BDD tests to improve coverage across 52 source files

Added 52 new .feature files and corresponding _steps.py files targeting
previously uncovered code paths in the following areas:

- TUI layer: app, commands, persona (state/schema/registry), widgets,
  input (shell_exec, reference_parser)
- Application services: plan lifecycle/service/executor, session,
  project, repo indexing, correction, checkpoint, actor, llm_actors,
  strategy coordinator, resource file watcher, service retry wiring
- CLI commands: session, resource, repl, plan, db, automation_profile
- Domain models: retry_policy, resource_type, cost_budget,
  docker_compose_analyzer, detail_level, _sql_string_aware,
  _postgresql_helpers
- Core: circuit_breaker, retry_service_patterns
- Infrastructure: repositories, transaction_sandbox, strategy_registry,
  plugins/loader, container
- Config: settings
- Agents: plan_generation, context_analysis, auto_debug
- A2A: facade

All new tests follow the Behave/Gherkin BDD standard. Resolved step
definition collisions with unique prefixes. Fixed Alembic fileConfig
logger disabling issue (disable_existing_loggers=False).

ISSUES CLOSED: #1068

features/auto_debug_coverage_boost.feature

@@ -0,0 +1,145 @@
+Feature: AutoDebug graph node coverage boost
+  Additional scenarios that exercise previously uncovered code paths
+  in the auto_debug.py graph module, targeting individual node methods
+  and edge-case LLM response handling.
+
+  Background:
+    Given the auto debug graph module is imported
+
+  # -----------------------------------------------------------------------
+  # _analyze_error node
+  # -----------------------------------------------------------------------
+  Scenario: Analyze error with Mock LLM response yields default analysis
+    Given an auto debug agent with a mock LLM returning "Mock LLM response"
+    When I call _analyze_error with an error state
+    Then the analysis message should be "Error analysis completed"
+
+  Scenario: Analyze error with real content passes through LLM output
+    Given an auto debug agent with a mock LLM returning "Root cause is a null pointer"
+    When I call _analyze_error with an error state
+    Then the analysis message should be "Root cause is a null pointer"
+
+  Scenario: Analyze error handles LLM invocation failure gracefully
+    Given an auto debug agent with a failing LLM
+    When I call _analyze_error with an error state
+    Then the analysis message should be "Error analysis completed"
+
+  # -----------------------------------------------------------------------
+  # _generate_fix node
+  # -----------------------------------------------------------------------
+  Scenario: Generate fix with Mock LLM response yields default fix data
+    Given an auto debug agent with a mock LLM returning "Mock LLM response"
+    And a state with an existing error analysis message
+    When I call _generate_fix with the prepared state
+    Then the current fix description should be "Fix suggestion"
+    And the current fix code should be "# Fixed code"
+
+  Scenario: Generate fix parses valid JSON from LLM
+    Given an auto debug agent with a mock LLM returning valid fix JSON
+    And a state with an existing error analysis message
+    When I call _generate_fix with the prepared state
+    Then the current fix description should be "Parsed JSON fix"
+    And the current fix code should be "print('fixed')"
+
+  Scenario: Generate fix falls back on invalid JSON from LLM
+    Given an auto debug agent with a mock LLM returning "not valid json at all"
+    And a state with an existing error analysis message
+    When I call _generate_fix with the prepared state
+    Then the current fix description should be "Fix attempt 1"
+    And the current fix code should be "not valid json at all"
+
+  Scenario: Generate fix handles LLM invocation failure gracefully
+    Given an auto debug agent with a failing LLM
+    And a state with an existing error analysis message
+    When I call _generate_fix with the prepared state
+    Then the current fix description should be "Fix suggestion"
+    And the current fix code should be "# Fixed code"
+
+  Scenario: Generate fix includes previous attempt count in fallback description
+    Given an auto debug agent with a mock LLM returning "still broken"
+    And a state with two previous attempted fixes
+    When I call _generate_fix with the prepared state
+    Then the current fix description should be "Fix attempt 3"
+
+  # -----------------------------------------------------------------------
+  # _validate_fix node
+  # -----------------------------------------------------------------------
+  Scenario: Validate fix with Mock LLM response marks fix as valid
+    Given an auto debug agent with a mock LLM returning "Mock LLM response"
+    And a state with a current fix to validate
+    When I call _validate_fix with the prepared state
+    Then the fix should be marked as validated
+
+  Scenario: Validate fix parses valid JSON with is_valid true
+    Given an auto debug agent with a mock LLM returning valid validation JSON true
+    And a state with a current fix to validate
+    When I call _validate_fix with the prepared state
+    Then the fix should be marked as validated
+
+  Scenario: Validate fix parses valid JSON with is_valid false
+    Given an auto debug agent with a mock LLM returning valid validation JSON false
+    And a state with a current fix to validate
+    When I call _validate_fix with the prepared state
+    Then the fix should not be marked as validated
+    And the current fix should be appended to attempted fixes
+
+  Scenario: Validate fix handles invalid JSON containing positive keywords
+    Given an auto debug agent with a mock LLM returning "The fix is valid and correct"
+    And a state with a current fix to validate
+    When I call _validate_fix with the prepared state
+    Then the fix should be marked as validated
+
+  Scenario: Validate fix handles invalid JSON without positive keywords
+    Given an auto debug agent with a mock LLM returning "The fix does not work"
+    And a state with a current fix to validate
+    When I call _validate_fix with the prepared state
+    Then the fix should not be marked as validated
+    And the current fix should be appended to attempted fixes
+
+  Scenario: Validate fix handles LLM invocation failure gracefully
+    Given an auto debug agent with a failing LLM
+    And a state with a current fix to validate
+    When I call _validate_fix with the prepared state
+    Then the fix should be marked as validated
+
+  # -----------------------------------------------------------------------
+  # _should_retry_fix routing
+  # -----------------------------------------------------------------------
+  Scenario: Should retry fix returns retry when validation failed and under max attempts
+    Given an auto debug agent with a mock LLM returning "Mock LLM response"
+    When I check should_retry_fix with validated false and 1 attempt
+    Then the routing decision should be "retry"
+
+  Scenario: Should retry fix returns done when validation succeeded
+    Given an auto debug agent with a mock LLM returning "Mock LLM response"
+    When I check should_retry_fix with validated true and 0 attempts
+    Then the routing decision should be "done"
+
+  Scenario: Should retry fix returns done when max attempts reached
+    Given an auto debug agent with a mock LLM returning "Mock LLM response" and max 2 fix attempts
+    When I check should_retry_fix with validated false and 2 attempts
+    Then the routing decision should be "done"
+
+  # -----------------------------------------------------------------------
+  # _finalize node
+  # -----------------------------------------------------------------------
+  Scenario: Finalize produces result dict from successful state
+    Given an auto debug agent with a mock LLM returning "Mock LLM response"
+    And a finalize state with fix_validated true and 0 attempted fixes
+    When I call _finalize with the prepared state
+    Then the result should indicate success true
+    And the result attempts count should be 0
+
+  Scenario: Finalize produces result dict from failed state
+    Given an auto debug agent with a mock LLM returning "Mock LLM response"
+    And a finalize state with fix_validated false and 3 attempted fixes
+    When I call _finalize with the prepared state
+    Then the result should indicate success false
+    And the result attempts count should be 3
+
+  # -----------------------------------------------------------------------
+  # Constructor edge case
+  # -----------------------------------------------------------------------
+  Scenario: AutoDebugAgent constructor raises ValueError when no LLM is provided
+    When I try to create an AutoDebugAgent with no LLM
+    Then a ValueError should be raised with the missing provider message