fix: preserve openai-compatible reasoning_effort option

[slegarraga]

Summary

Addresses the snake_case provider-option path from #75.

@ai-sdk/openai-compatible request construction filtered reasoning_effort out of provider options because only reasoningEffort was in the parsed schema, then wrote reasoning_effort from the camelCase value. That made an explicit reasoning_effort: "none" no-op for OpenAI-compatible endpoints such as Ollama's /v1/chat/completions.

This change:

• accepts reasoning_effort in the openai-compatible provider options schema
• writes the OpenAI-compatible request body from reasoning_effort ?? reasoningEffort
• adds request-body regression coverage showing snake_case wins when both forms are present

Validation

• bun test test/provider/copilot/copilot-chat-model.test.ts
• bun run typecheck
• git diff --check

[aictrl-dev]

Review notes (not line-anchored)

• [CONSISTENCY] Incomplete test coverage for reasoning_effort variants — The test only covers the case where both reasoningEffort and reasoning_effort are provided simultaneously, asserting snake_case priority. It does not verify the fallback path where only reasoningEffort (camelCase) is provided — if the implementation is supposed to normalize camelCase to snake_case, that path is untested and could silently break. Likewise, the standalone reasoning_effort (snake_case only) case is not covered. At minimum, add two tests: one passing only reasoningEffort: 'high' asserting the expected body key/value, and one passing only reasoning_effort: 'none' to confirm it passes through unchanged.

[aictrl-dev]

Now let me check sibling files for consistency patterns before dispatching per-file subagents.Let me dispatch the per-file subagents now.Now let me save the findings and merge them.Now let me record the review and submit it.Now let me record the findings and complete the review.2 inline comment(s) + 1 PR-level note(s) posted on PR #76.

Findings summary:

Duplicate Code

• openai-compatible-chat-options.ts:20 — reasoning_effort duplicates the existing reasoningEffort field at line 15. Both are z.string().optional() representing the same concept, breaking the all-camelCase naming convention in this schema. Consider normalizing via .transform() into a single canonical field.
• openai-compatible-chat-language-model.ts:175 — The ?? fallback (reasoning_effort ?? reasoningEffort) deviates from the single-source rename pattern used for other fields like verbosity: compatibleOptions.textVerbosity.

Test Coverage Gap (body note)

• The new test only covers the case where both fields are provided simultaneously. Missing coverage for camelCase-only fallback and snake_case-only passthrough paths.

[byapparov]

Test coverage gap addressed in 1494c22: added snake_case-only and camelCase-only cases alongside the existing both-set precedence test, and documented the alias/precedence relationship in the options JSDoc. The .transform() normalization suggested in the inline comments does not work here — see thread replies for the two concrete failure modes (loss of .shape used by the passthrough filter, and undefined-key clobbering in the Object.assign namespace merge).