Add session-aware dirty-files tracking

[severity1]

Summary

Support session-isolated dirty-file tracking to prevent interference between concurrent Claude sessions editing the same project.

Problem

Currently, dirty files are tracked in a single shared file: .claude/auto-memory/dirty-files

When multiple Claude sessions run concurrently on the same project:

1. Session A edits file1.py -> appends to dirty-files
2. Session B edits file2.py -> appends to dirty-files
3. Session A's stop hook triggers memory update for both files (including B's changes)
4. Session B may see empty dirty-files (if A cleared it) or trigger duplicate processing

This causes:

• Sessions processing files they didn't edit
• Race conditions on file reads/writes
• Duplicate or missed memory updates
• Unpredictable behavior in multi-session workflows

Proposed Solution

Use session-specific dirty-file tracking:

.claude/auto-memory/
  dirty-files-{session_id_1}
  dirty-files-{session_id_2}
  ...

Implementation steps:

1. Check if Claude provides a session identifier via environment variable (e.g., CLAUDE_SESSION_ID)
2. Modify post-tool-use.py to write to session-specific file
3. Modify stop.py to read only the current session's dirty file
4. Add cleanup for stale session files (sessions that terminated without cleanup)

Considerations

• Session ID availability: Need to verify Claude exposes a session identifier
• Stale file cleanup: Sessions may crash without clearing their dirty files
• Backwards compatibility: Handle upgrade path from single-file to session-aware format

Affected Files

• scripts/post-tool-use.py - dirty file writes
• scripts/stop.py - dirty file reads
• agents/memory-updater.md - may need session context