fix: sync validNames with ProviderName type + handle provider-prefixed model IDs (OpenRouter)

[nalepy]

Summary

Two small fixes for provider management in the web dashboard.

Fix 1 — ValidNames array out of sync with ProviderName type

File: src/web/api/providers.ts

The validNames whitelist in the POST /api/providers/:name handler was missing 5 provider types that exist in the ProviderName union type: openaiCompat, mimo, mimoTokenPlan, chatgptWeb, githubCopilot. Toggling any of these providers from the web UI returned 400 Unknown provider.

Fix: Added the 5 missing names to the validNames array. Now matches ProviderName exactly.

Fix 2 — isOpenAIChatModel rejects provider-prefixed model IDs

File: src/utils/provider-models.ts

isOpenAIChatModel() filtered models by checking id.startsWith("gpt-"). OpenRouter (and similar API proxies) return model IDs with a provider prefix: openai/gpt-4o-mini, anthropic/claude-sonnet-4, etc. The prefix caused ALL models to be filtered out, producing the error "Mercury could not find any supported chat models for this provider."

Fix: Split the model ID on / and check only the last segment. This handles both formats:

• OpenRouter: openai/gpt-4o-mini → gpt-4o-mini ✓
• Direct OpenAI: gpt-4o-mini → gpt-4o-mini ✓
• OpenAI o-series: o3-mini → o3-mini ✓

Non-OpenAI models from proxy services (e.g. anthropic/claude-sonnet-4) are still correctly excluded because claude-sonnet-4 doesn't match the gpt-* or /^o\d/ filters.

Testing

• mercury doctor → toggle MiMo provider no longer returns 400
• mercury doctor → Test Connection with OpenRouter's OpenAI endpoint now returns models
• Direct OpenAI API still works (backward compatible)
• Non-OpenAI proxy models still correctly excluded

Files changed

File	Change
src/web/api/providers.ts	Add 5 missing provider names to validNames
src/utils/provider-models.ts	Split model ID on / before prefix check