fix: remove misdirected SemanticMsg enqueue causing O(n²) reprocessing

[chethanuk]

Summary

• Removed misdirected SemanticMsg enqueue in session.py (lines 309-324): After _flush_semantic_operations() already enqueued correct messages with change-tracking dicts, commit_async() was enqueuing a second SemanticMsg targeting the session URI (not a memory directory) with changes=None
• Hardened cache loading in semantic_processor.py: _process_memory_directory() now always loads cached summaries from .overview.md regardless of msg.changes value (defense-in-depth against other changes=None callers like lock_manager.py and summarizer.py)
• Added 2 regression tests verifying both the enqueue removal and the cache guard

Problem

On every commit, session.py:309-324 enqueued a SemanticMsg with changes=None after the compressor had already correctly enqueued messages with proper change-tracking dicts. This misdirected message bypassed the cache guard in _process_memory_directory() (which only loaded cached summaries when msg.changes was truthy), causing unconditional VLM API calls for every memory file — O(n²) reprocessing. At 500 memories this wasted 100K+ tokens/day.

The timeline on each commit

1. commit_async() starts
2. → compressor.extract_long_term_memories()
3. → compressor._flush_semantic_operations()
4. → enqueues SemanticMsg(uri=memory_dir, changes={added: [...]}) ✓ CORRECT
5. → session.py lines 309-324
6. → enqueues SemanticMsg(uri=session_dir, changes=None) ✗ MISDIRECTED
7. Processor handles msg from step 4: loads cache, only regenerates changed files
8. Processor handles msg from step 6: no cache, VLM call for EVERY file

Step 6-8 is waste. Removing lines 309-324 eliminates step 6, so step 8 never happens.

Fix

1. session.py: Removed 16-line block that enqueued the misdirected SemanticMsg — the compressor's _flush_semantic_operations() already handles this correctly
2. semantic_processor.py: Removed if msg.changes: guard on cache loading so .overview.md is always consulted
   Closes Memory extraction triggers O(n²) semantic reprocessing — token cost grows quadratically with memory count #505

Testing

• 2 regression tests added (test_fix_505_duplicate_semantic_enqueue.py)
  • test_no_misdirected_semantic_enqueue_after_flush — verifies commit_async() doesn't enqueue after compressor flush
  • test_process_memory_directory_loads_cache_when_changes_none — verifies cache is loaded even when msg.changes is None
• Both tests confirmed RED before fix, GREEN after fix (TDD cycle)
• Full test suite: no regressions (pre-existing config errors unrelated)

Related Issue

Type of Change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Refactoring (no functional changes)
• Performance improvement
• Test update

Changes Made

Testing

• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• I have tested this on the following platforms:
  • Linux
  • macOS
  • Windows

Checklist

• My code follows the project's coding style
• I have performed a self-review of my code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• Any dependent changes have been merged and published

Screenshots (if applicable)

Additional Notes