fix(tui,cli,tests): harden persona/input modes and stabilize parallel test execution
- Add shell safety controls for REPL/TUI (`looks_dangerous`, confirmation gate, timeout handling, and env-based shell disable guard).
- Secure persona workflows with strict name/path validation, safe import/export resolution, atomic+locked registry writes, and malformed YAML resilience.
- Unify persona models and wiring by reusing canonical TUI schema/registry, adding DI providers, and lazy-loading TUI exports to avoid circular imports.
- Improve reference discovery with ignored-directory filtering, symlink-safe walking, and TTL caching for CLI/TUI reference catalogs.
- Expand Behave/Robot coverage for safety/error paths and parallel-isolation behavior; add shared `features/mocks/fake_repl_input.py` helper.
- Fix parallel-run flakiness via deterministic cleanup/reload patterns and watchdog polling fallback when inotify limits are reached.
ISSUES CLOSED: #695
aditya
d9e51d98f8
