chore: revert "feat: pass invocation_state to edge condition calls (#2305)"

[zastrowm]

Summary

Reverts #2305.

This PR introduced a regression that breaks all Graph integration tests that use FileSessionManager with a TypeError: Object of type Agent is not JSON serializable.

Root Cause

The PR stored self._current_invocation_state in serialize_state() and passed the same dict reference to agent node executors. Python's Agent._execute_event_loop_cycle mutates the passed invocation_state dict by adding invocation_state["agent"] = self (a backwards-compat shim in agent.py:1012). After a node completes, the session manager hook fires sync_multi_agent → serialize_state() → json.dump(), which hits the non-serializable Agent object now sitting in self._current_invocation_state.

Why revert rather than fix-forward

1. We're attempting to get a release out today
2. Want to revisite whether or not persistence makes sense here

[codecov]

Codecov Report

❌ Patch coverage is 77.77778% with 2 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
strands-py/src/strands/multiagent/graph.py	75.00%	1 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!

[github-actions]

Assessment: Approve

Clean, well-motivated revert that fixes a serialization regression in Graph integration tests. The PR description provides excellent root cause analysis and clear rationale for reverting rather than fixing forward.

Review Notes

• Correctness: The three architectural concerns (dead-on-arrival persisted value, TypeScript parity, unintended serialization constraint) justify reverting over a quick fix.
• Scope: Changes are limited to removing the feature introduced in feat: pass invocation_state to edge condition calls #2305 — no unrelated modifications.
• Testing: Removed tests correspond exactly to the removed feature; existing test suite remains intact.
• API surface: EdgeCondition and EdgeConditionWithContext were introduced in the same release being reverted, so removal carries no backwards-compatibility risk.

The acknowledgement that the feature itself is valuable (edge conditions receiving invocation_state) but needs a design that avoids persisting or sharing mutable references is a good callout for future work.