docs(reference): fix invariant precedence chain to show 4-tier (plan > action > project > global)

The reference doc showed a 3-tier chain (plan > project > global) which
was inconsistent with the main specification and implementation.

The correct 4-tier chain is: plan > action > project > global.
This matches docs/specification.md line 92 and the implementation's
_SCOPE_PRECEDENCE dict in actor/reconciliation.py.

Refs: #5861

docs/reference/invariants.md

@@ -18,13 +18,13 @@ Actor before any strategy work begins.
 When computing the effective set of invariants for a plan, the precedence chain is:
 
 ```
-plan > project > global
+plan > action > project > global
 ```
 
 - **Plan-level** invariants take highest precedence.
+- **Action-level** invariants apply next. Action invariants are also promoted to plan scope when `plan use` is called, but they remain resolvable at action scope during reconciliation.
 - **Project-level** invariants apply next.
 - **Global-level** invariants are lowest precedence.
-- **Action-level** invariants are promoted to plan scope when `plan use` is called.
 
 ### De-duplication