[merge-gateway] Add reasoning options

[rekram1-node]

Summary

• backfill explicit reasoning options for all 59 reasoning-capable Merge Gateway routes
• expose Merge's documented Anthropic extended-thinking controls on eight supported Claude routes
• retain [] where Merge does not publish a selectable control or provider-family mapping

Evidence standard

• Merge documents /v1/openai and /v1/anthropic as translation adapters, not raw-body proxies: https://api-gateway.merge.dev/v1/openapi.json
• The live native ResponseRequest documents top-level thinking and free-form provider_options, but publishes no provider-family schemas for Gemini, GLM, DeepSeek, Mistral, xAI, or OpenAI reasoning controls.
• Merge's official Gateway feature guide documents extended thinking specifically for Anthropic Claude: https://github.com/merge-api/merge-gateway-skills/blob/main/skills/gateway-features/SKILL.md
• Merge's public model catalog identifies model availability but does not publish per-model reasoning controls: https://docs.merge.dev/merge-gateway/api-overview/models/list
• Unknown fields and additionalProperties establish parser tolerance, not translation or upstream propagation. Native model capability alone is not treated as a Merge control.

Result

• 8 configurable Claude routes with toggle plus bounded manual budgets
• 51 fixed/no-Merge-documented-control routes
• [] means no verified selectable control through Merge, not that the model lacks reasoning

Validation

• bun validate
• git diff --check
• resolved coverage: 59/59 explicit
• no options on non-reasoning models
• no unbounded token budgets

[rekram1-node]

Superseded by 9 provider/lab-scoped replacement PRs, all validated and limited to at most 15 model files:

#2302 #2303 #2304 #2305 #2306 #2307 #2308 #2309 #2310