Harness 3.2 Model picker respects the harness [both repos] #6544

Part of #6535.

• Today a GPT model on a Claude Code client passes preflight and fails after the box boots (BB-245).
• Filter pickers by harness using the support table.
• Refuse incompatible models when a client or environment is saved, with an actionable error. The places to add this:
  • ensureHostConfigV2 (convex/lib/hostConfigV2.ts)
  • validateLaunchableModel (convex/projectEnvironments.ts)
• Make the eval dry run check the environment's model override (checkEvalHarnessStaticAdmission, server/routes/v1/evals.ts ~4916).
• Regression check only: assistant-turn.ts (~675-705) already refuses an incompatible harness model instead of running the emulated engine. Add or keep a test that pins this behavior.

Delivery rules

• Fresh worktree and branch; each PR at most 400 handwritten changed lines; stack when necessary.
• Verify the actual failing boundary and run repository checks.
• Preserve broker-only model auth, no venue/engine fallback, and separate infrastructure failures.
• Human-only: production deployment, flags/env changes, rebuilds/backfills, signed/published packs and large paid runs.

Status: planned; completion requires the acceptance evidence above.

Pull request stack

• MCPJam/mcpjam-backend#2197: fix(harness): reject incompatible models when saving clients

The implementation hub records merge order, current verification and remaining release gates. PR publication does not mark the deployed or packaged acceptance matrix complete.