Fix DeepSeek V4.1 reasoning effort catalog

[razorback4417]

deepseek-v4.1-flash rejects medium, but setup currently advertises low/medium/high for every Webster model and defaults to medium. That leaves Codex selecting an effort the backend won't accept.

This adds one catalog override for V4.1:

• default: high
• supported: low, high, xhigh, max

All other discovered models keep the current low/medium/high profile, and the proxy still forwards requests unchanged.

I kept this in catalog generation because Webster's /v1/models response doesn't expose reasoning capabilities. We shouldn't need an entry for every new model—only models whose contract differs from the shared default. If Webster starts returning capability metadata, that should replace the local overrides.

Tested with the proxy suite and the repository test suite.