PR C: Switch default model to gemini-flash (conditional on spike #43)

[jakebromberg]

Problem

Conditional follow-up to the Flash calibration spike (#43). If Flash meets the pre-committed quality threshold recorded in that spike, the default Gemini model in core/gemini.py swaps from Pro to Flash. Most of the corpus spend gets a 5-10× cost cut for whatever quality delta the spike accepted.

End state

GeminiClient defaults to the Flash model ID. Pro stays available as an explicit constructor argument so it can be used for selective re-extraction (Sprint 3 might do this) or one-off comparisons.

Where

• core/gemini.py — change the default model ID.
• tests/unit/test_gemini.py — update the model-property test.
• CLAUDE.md — mention the default swap with the spike's calibration score as justification.
• scripts/calibrate_models.py adapter, if the spike landed it in tree.

Constraints

• Do NOT merge this PR unless the spike (Spike: 5-page Gemini Flash calibration #43) PASSES its pre-committed quality threshold. If the spike fails, close this issue without merging; Sprint 1's cost wins are what we keep.
• This is a single-line model-ID change — keep the diff minimal. No prompt edits, no schema edits.

Acceptance criteria

• Spike Spike: 5-page Gemini Flash calibration #43 has recorded a PASS verdict.
• Default model swapped in core/gemini.py.
• Test for GeminiClient.model property updated.
• CLAUDE.md notes Flash as the corpus default and cites the spike's score.
• 5-golden calibration confirms the post-merge score matches what the spike measured.

Related

• Sprint 2 parent: Sprint 2: Gemini Flash tier decision #38
• Blocked by spike: Spike: 5-page Gemini Flash calibration #43