Fix artifact edit reliability and raise output limits

[iplusplus-nl]

What changed

• Rebase independent artifact patches that all reference the original source while continuing to reject conflicting overlaps.
• Ask edit models for atomic, bounded patch batches and disable unnecessary reasoning on OpenRouter-backed edit calls.
• Surface output-token exhaustion with an actionable message.
• Use one shared 32,000-token ceiling for Responses, Chat Completions, browser-direct generation, and artifact editing.

Why

Production logs showed two separate failures: one model returned a stale 21st patch after earlier patches changed its larger source context, and another exhausted the managed service's 16,000-token ceiling on hidden reasoning. The former discarded an otherwise usable edit batch; the latter surfaced as a generic 502.

Impact

Artifact edits can safely recover independent original-source patch batches, edit-only calls do not waste their output budget on hidden reasoning, and both ordinary generation and editing can return up to 32,000 tokens.

Validation

• npm test — 1,208 tests passed
• npm run build — production TypeScript and Vite build passed