fix: show standard prices in provider model listings

[chrarnoldus]

Summary

• Preserve frontend endpoint variant metadata and exclude batch cards before provider snapshots are deduplicated.
• Keep standard pricing independent of card order while preserving free variants and older metadata shapes.
• Add regression coverage at the upstream fetch/parse boundary, including GPT-5.6 Sol prices.

Root cause

The Providers & Models screen reads the synced frontend provider cards, not the public model endpoints directly. OpenRouter currently returns a standard GPT-5.6 Sol card followed by a batch card with the same base model slug. Variant metadata was stripped, so last-write-wins deduplication replaced standard prices with batch prices (which match the flex prices in the linked public endpoints response). This also affected other standard/batch pairs.

The fix leaves discount reversal and public endpoint routing unchanged. The corrected screen prices appear after the next successful provider metadata sync.

Reference: https://openrouter.ai/api/v1/models/openai/gpt-5.6-sol/endpoints

Verification

• Passed formatting and focused oxlint on all four changed files.
• Passed pnpm --filter @kilocode/db typecheck.
• pnpm --filter web typecheck could not complete: the two-minute sandbox limit terminated the prerequisite tRPC build. Local Node is 22 rather than the required 24.
• Tests intentionally deferred to CI as requested.
• Focused independent code review: no findings.

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

Files Reviewed (4 files)

• apps/web/src/lib/ai-gateway/providers/openrouter/fetch-provider-models.test.ts
• apps/web/src/lib/ai-gateway/providers/openrouter/fetch-provider-models.ts
• apps/web/src/lib/ai-gateway/providers/openrouter/sync-providers.ts
• packages/db/src/schema-types.ts

---

_{Reviewed by grok-4.6 · Input: 59.1K · Output: 13.6K · Cached: 428.3K}

_{Review guidance: REVIEW.md from base branch main}