feat(openai-codex): add gpt-5.3-codex-spark model metadata

[aShanki]

Summary

• add gpt-5.3-codex-spark to OpenAI Codex model list
• set Spark-specific metadata (contextWindow: 128000, maxTokens: 8192, supportsImages: false)
• add model coverage in OpenAI Codex provider tests

Testing

• pnpm vitest run api/providers/__tests__/openai-codex.spec.ts (in src)
• pnpm --filter @roo-code/types test -- src/__tests__/provider-settings.test.ts

Start a new Roo Code Cloud session on this branch

[roomote]

PR Review: feat(openai-codex): add gpt-5.3-codex-spark model metadata

Verdict: Sound

This PR does exactly what it says -- adds a single new model entry (gpt-5.3-codex-spark) to the OpenAI Codex provider and covers it with tests. No scope creep detected.

---

Scope Check

• Files changed: 2 (model definition + test). Perfectly scoped.
• No unrelated rewrites: The only modification to existing code is appending "gpt-5.3-codex-spark" to the it.each array in the test file.
• No scope creep: No refactors, no renames, no tangential changes.

Contract Consistency

All fields in the new model entry are valid members of modelInfoSchema (from packages/types/src/model.ts). Verified:

• maxTokens, contextWindow, includedTools, excludedTools, supportsImages, supportsPromptCache, supportsReasoningEffort, reasoningEffort, inputPrice, outputPrice, supportsTemperature, description -- all present in the Zod schema.
• includedTools: ["apply_patch"] and excludedTools: ["apply_diff", "write_to_file"] are consistent with every other model in the file.
• inputPrice: 0 / outputPrice: 0 follows the subscription-based pricing convention.

Intentional Differentiation from Other Models

The Spark variant intentionally differs from other Codex models:

• contextWindow: 128000 (vs 400000 for others) -- smaller context for a fast model.
• maxTokens: 8192 (vs 128000 for others) -- significantly lower output cap.
• supportsImages: false (vs true for others) -- text-only, matching the description.

These differences are internally consistent and align with the "fast, text-only" positioning described in the model description.

Test Coverage

• Added to the parameterized it.each test for valid model IDs.
• Dedicated test asserting Spark-specific limits (contextWindow, maxTokens, supportsImages).
• CI passes on all platforms (ubuntu + windows).

Findings

No critical or high-severity issues found.

Severity	Finding
P3 (Low)	Missing // Subscription-based: no per-token costs comment above inputPrice: 0 that most other model entries include. Purely cosmetic -- some entries also omit it (e.g., gpt-5.3-codex, gpt-5.2-codex), so this is consistent with the newer entries.
P3 (Low)	No .changeset file included. If this repo requires changesets for releases, one may need to be added before merge. CI does not appear to enforce this for non-bump PRs, so likely fine.

Summary

Clean, minimal, well-tested addition. Approved.

[damianar1984]

@hannesrudolph could you also approve and merge #11479