v0.5.1 — synonyms + CI guard upgrade (combined corpus 95% coverage)

Patch release rolling up two iterations since v0.5.0:

What's new

• Synonym overrides (#68): closes the clean ↔ clear gap exposed by blind-v2. Answerable coverage 80% → 90% on blind-v2; combined corpus 90% → 95% (19/20).
• CI guard upgrade (#67): regression workflow now checks both v1 (threshold 0.95) AND blind-v2 (threshold 0.75). Either regression blocks the PR.

Compatibility

• No protocol changes.
• No schema changes.
• All v0.5.0 redb stores remain back-compatible.

Benchmark headline

Corpus	v0.5.0	v0.5.1	Δ
v1 audited (13q)	100%	100%	parity
blind-v2 (15q)	80%	90%	+10pp
combined (28q)	90%	95%	+5pp

"what cleans up after analysis?" — previously a miss — now hits at rank 0 with top1=clear thanks to the new clean ↔ clear synonym override.

See CHANGELOG.md for the full per-PR breakdown.

What's Changed

• ci: add blind-v2 corpus to the regression guard (threshold 0.75) by @njfio in #67
• feat(rts-bench): stemmer synonym overrides — closes clean↔clear gap (+10pp on blind-v2) by @njfio in #68
• release: v0.5.1 — patch (synonym overrides + CI guard upgrade) by @njfio in #69

Full Changelog: v0.5.0...v0.5.1

What's Changed

• ci: add blind-v2 corpus to the regression guard (threshold 0.75) by @njfio in #67
• feat(rts-bench): stemmer synonym overrides — closes clean↔clear gap (+10pp on blind-v2) by @njfio in #68
• release: v0.5.1 — patch (synonym overrides + CI guard upgrade) by @njfio in #69

Full Changelog: v0.5.0...v0.5.1

What's Changed

• ci: add blind-v2 corpus to the regression guard (threshold 0.75) by @njfio in #67
• feat(rts-bench): stemmer synonym overrides — closes clean↔clear gap (+10pp on blind-v2) by @njfio in #68
• release: v0.5.1 — patch (synonym overrides + CI guard upgrade) by @njfio in #69

Full Changelog: v0.5.0...v0.5.1