feat(mitm): add Antigravity reasoning-effort overrides

[diegosouzapw]

Summary

• Antigravity MITM alias entries now support an optional reasoning-effort override alongside the existing destination-model mapping ({ model?, reasoningEffort? }); legacy plain-string mappings still normalize to { model } with no DB migration required.
• The standalone Antigravity proxy forwards the chosen tier as a top-level reasoningEffortOverride; the antigravity→openai translator honors it ahead of its thinkingConfig-derived guess, and an explicit "none" suppresses reasoning_effort even when Antigravity's own request asked for thinking.
• The Antigravity tool card now exposes a per-model reasoning-effort selector (Default/None/Low/Medium/High/XHigh) next to the model-mapping input, reusing OmniRoute's existing canonical 5-tier reasoning vocabulary instead of introducing a new one.

Attribution

Thanks to @trfi for the original implementation.

Changes

• src/mitm/aliasConfig.ts (new) + src/mitm/_internal/aliasConfig.cjs (new, CJS mirror for the standalone proxy process): normalize/validate alias entries and reasoning-effort values.
• src/mitm/server.cjs: getMappedModel → getMappedOverride; intercept() applies the override (model swap + reasoningEffortOverride) via the new shim.
• open-sse/translator/request/antigravity-to-openai.ts: honor an explicit reasoningEffortOverride ahead of the thinkingConfig-derived tier.
• src/app/api/cli-tools/antigravity-mitm/alias/route.ts + src/shared/validation/schemas/cli.ts: accept the structured mapping shape, validate reasoning-effort values at the API boundary.
• src/app/(dashboard)/dashboard/cli-code/components/AntigravityToolCard.tsx: per-model reasoning-effort selector.
• i18n: new cliTools.reasoningEffort* keys in en.json + placeholder entries in all other locales.

Test plan

• TDD: tests/unit/mitm-antigravity-reasoning-effort-override.test.ts (14 cases) and tests/unit/mitm-alias-config-shim.test.ts (6 cases) — confirmed failing against the pre-change translator (override ignored: low !== high, none did not suppress reasoning_effort, reasoning-only override was a no-op) and passing after the fix.
• npm run typecheck:core — clean.
• npx eslint on all changed files — clean (the one pre-existing AntigravityToolCard.tsx react-hooks/exhaustive-deps warning was confirmed present on the unmodified base file, not introduced by this change).
• Existing tests/unit/translator-antigravity-to-openai.test.ts (9 cases) and tests/unit/antigravity-orphan-toolresult-6026.test.ts (3 cases) still pass — no regression.

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[diegosouzapw]

Aprovado. Diff grande mas escopo bate 1:1 com o body (i18n é churn mecânico de placeholder MISSING em 44 locais, padrão já usado no projeto). Revert-proof no translator confirma fail-without-fix (2/14 testes caem sem o fix). Validação de segurança OK: Hard Rule #13 (sem interpolação de shell no .cjs) e #7 (Zod + hasInvalidReasoningEffort rejeitando valor de reasoning inválido com 400 na borda da API). Único ponto de atenção não-bloqueante: a duplicação aliasConfig.ts/aliasConfig.cjs é uma dívida arquitetural inerente (server.cjs não pode importar ESM/TS) — os dois têm teste próprio, mas nada garante que fiquem sincronizados se só um for editado no futuro; vale considerar um gerador ou um teste cross-check de paridade entre os dois módulos numa PR futura. Pronto para merge como está.