chore(models): migrate dated snapshots to stable aliases

[silversurfer562]

Summary

Mechanical migration: claude-haiku-4-5-20251001 → claude-haiku-4-5 in 3 sites — reranker default model, expander default model, probe_v3 script.

Companion to attune-ai PR #580 and attune-help PR #11. Phase 0.2 of the family plan.

UX-guard

Spine-repo change — extra care taken:

• Pure identifier swap. Stable alias routes to the current checkpoint in the Haiku 4.5 series. Same model behavior expected.
• Public-API surface impact: the model= default arg on LLMReranker and QueryExpander constructors. Consumers passing an explicit model string are unaffected.
• Defensive evidence: 920 unit tests pass locally. Golden tests deselected from local run (require API spend); CI will exercise them.
• Per the family plan UX-guard: no retrieval / ranking / faithfulness behavior change is mechanistically possible from an ID swap. The guard's intent (catch synthetic-wins-but-real-corpora-lose regressions) doesn't apply here.

Test plan

• 920 unit tests pass
• CI matrix green on push (including golden tests)
• Perf-gate green

🤖 Generated with Claude Code

[github-actions]

Perf delta — within baseline

Blocking on regression in: keyword_retriever_retrieve.cpu, rag_pipeline_run.cpu. Other metrics are advisory and don't block merge.

Metric	Baseline mean (s)	Current mean (s)	Δ	Threshold (s)	Status
directory_corpus_load.cpu	0.000055	0.000053	-3.6%	0.000059	ok
directory_corpus_load.wall	0.000054	0.000053	-1.9%	0.000058	ok
keyword_retriever_retrieve.cpu	0.005437	0.003772	-30.6%	0.005631	ok
keyword_retriever_retrieve.wall	0.005437	0.003772	-30.6%	0.005632	ok
rag_pipeline_run.cpu	0.000633	0.000557	-12.0%	0.000793	ok
rag_pipeline_run.wall	0.000632	0.000557	-11.9%	0.000793	ok

[github-actions]

Downstream attune-gui — green

Blocking gate (Phase 4 W3.2 onwards). Failure here fails the job and blocks merge.

Field	Value
attune-rag PR SHA	4a209ade01367f3118b9a656f0dd4157a83ef903
attune-gui ref	feature/attune-rag-0.2-editor-rename
Test selector	pytest sidecar/tests -k 'editor or rag'
Status	✅ pass
Workflow run	link

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!