Wynelson94/longhand

v0.8.0: cleaner narratives + canary suite + recall validator

Half the recall narratives in a typical user's corpus get cleaner
immediately on upgrade. The bug-finding test layer underneath means the
fixes that landed this release don't come back, and the next round of
bugs is cheaper to pin.

User-facing fixes: stripped leaked "Intent:" label from fix_summary
(migration v4 cleans existing rows), whitespace-boundary truncation
of fix_summary diffs (no more mid-token cuts), narrative footer now
shows session_id for "Other matches" lines, recall surfaces secondary
segment matches when episodes win primary.

Test infrastructure: canary harness for real-bug regression tests,
real-corpus recall validator (scripts/recall_diff.py).

Quiet simplification: removed redundant event_semantic_boost ranking
signal and its supporting vector-search call.

205 tests passing.