Auto-discover models from provider /v1/models endpoint

[djgpp6]

Fixes #204

What

Adding a custom model in Shelley currently means typing the model name by hand. A lot of providers (and proxies) only expose one model, or have a small set you can list via /v1/models. This wires that up so the model field fills itself.

Changes

• Add POST /api/custom-models-discover to fetch models from the remote provider (OpenAI-style /v1/models, Anthropic /v1/models, Gemini /v1beta/models)
• In the custom model form, discover models once endpoint + API key are set (debounced)
• Merge discovered models into the autocomplete list; auto-select when there is exactly one
• Unit tests for URL derivation and response parsing

Testing

• go test ./server/... -run 'TestModelsListURL|TestDiscover|TestHandleDiscover'

[cla-bot]

We require contributors to sign our Contributor License Agreement, and we don't have you on file. In order for us to review and merge your code, please contact @crawshaw at david@bold.dev to get yourself added.

[djgpp6]

We require contributors to sign our Contributor License Agreement, and we don't have you on file. In order for us to review and merge your code, please contact @crawshaw at david@bold.dev to get yourself added.

i think i did sign the CLA @philz

[philz]

@djgpp6 We can't seem to find your CLA (or maybe are having trouble associating usernames). E-mail philip@bold.dev and david@bold.dev and we'll get it sorted out?

[djgpp6]

yea mayhaps it's because i changed my username (previous one was okottorika). nonetheless, i'll mail you again

[philz]

Are you still interested in this? Sorry I dropped the ball for a while on these. We've got a fair bit of auto-discovery against the llm integration now, which would need to be somewhat unified perhaps.