feat(venice): add reasoning options

[rekram1-node]

Summary

• add audited Venice reasoning_options without treating /models capability metadata as authoritative over curated evidence
• rebuild the Venice generator around documented override > existing curated options > catalog options for uncurated new models
• report docs/catalog or curated/catalog discrepancies instead of overwriting curated controls
• leave catalog-false models undefined unless an explicit documented or curated [] establishes fixed reasoning

Sources and conflicts

• Venice guide: https://docs.venice.ai/guides/features/reasoning-models
• Venice /models metadata currently conflicts with the guide for Claude 4.5/4.6, GPT-5.2/Codex, Gemini Flash, Kimi K2.5, Qwen 3.5 35B, and GLM 5.1
• model-specific guide-backed overrides are documented in generate-venice.ts
• GPT-5.4/5.5 values are provisional first-party-backed overrides pending funded Venice probes:
  • https://developers.openai.com/api/docs/models/gpt-5.4
  • https://developers.openai.com/api/docs/models/gpt-5.4-pro
  • https://developers.openai.com/api/docs/guides/reasoning

Generator policy

• explicit documented override wins over existing and catalog values
• existing curated reasoning_options, including [], survive stale or false catalog claims
• catalog options populate only truly new models (existing === null) when supportsReasoningEffort=true and options are present; existing unresolved TOMLs remain undefined
• supportsReasoningEffort=false alone never manufactures []
• schema fields remain optional to tolerate incomplete catalog responses
• drift output reports exact option differences and preservation discrepancies
• Venice compute-disable toggles are not conflated with the none effort value

Provisional unresolved models

No Venice-specific effort proof was found for these current endpoints, so the PR intentionally leaves reasoning_options undefined rather than asserting fixed reasoning:

• Claude Opus 4.7 / 4.7 Fast / 4.8 / 4.8 Fast
• DeepSeek V4 Flash / V4 Pro
• Grok 4.20 / 4.20 Multi-Agent / 4.3 / Build 0.1
• MiniMax M3
• Qwen 3.6 Plus / 3.7 Plus / 3.7 Max

These require funded Venice inference probes. The prior probe budget was exhausted, so no new inference requests were made for this correction.

Verification

• bun test packages/core/test/venice-generator.test.ts: 8 passed, 0 failed
• env -u VENICE_API_KEY bun run venice:generate --dry-run: no writes; corrected mappings preserved; 17 catalog discrepancies reported; only 15 unrelated new catalog models and 2 unrelated pricing updates proposed
• bun validate: passed
• git diff --check: passed
• complete PR diff reviewed; no secrets, generated artifacts, or unrelated model writes included

This PR remains draft pending funded live verification of provisional and unresolved controls.