feat(ovhcloud): add reasoning options

[rekram1-node]

Summary

• audit all 15 active OVHcloud chat models against the official OVHcloud catalog
• add exact reasoning controls to all 6 reasoning models: Qwen3-32B toggle, gpt-oss low/medium/high effort, and three fixed/non-configurable declarations
• preserve authored reasoning options when reasoning remains enabled, while removing stale options when a translator disables reasoning
• cover OVHcloud translation and shared end-to-end sync behavior with focused regression tests

Verification

• bun test packages/core/test/ovhcloud-sync.test.ts packages/core/test/sync-runner.test.ts (12 passed)
• broader synced-provider tests for Baseten, Cloudflare Workers AI, Google, Venice, xAI, and Vercel (20 passed)
• bun validate
• bun models:sync ovhcloud --dry-run (15 unchanged)
• git diff --check

Audit notes

• OVHcloud documents /no_think for Qwen3-32B, represented as a toggle.
• OVHcloud documents low, medium, and high reasoning effort for both gpt-oss models.
• OVHcloud documents no configurable controls for the remaining three reasoning models, so they use empty options.
• No OVHcloud API credential was available for authenticated probes; the public catalog endpoint was queried live.