API gateway model id uses a single-colon `providerId:modelId` separator

What changed

When calling the local API gateway, the model field is now parsed as providerId:modelId, split on the first :. The v1 server split the provider and model on a double-colon ::. So a request that used to send "model": "openai::gpt-4o" must now send "model": "openai:gpt-4o"; the provider id is everything before the first colon, the model id is everything after it (so a model id may itself contain colons).

Why this matters to the user

Any external client, script, or agent that points at Cherry Studio's gateway (/v1/chat/completions, /v1/messages, /v1/responses) and hard-codes the provider::model form will now fail model resolution — the whole provider::model string is treated as the provider id and the leading : makes the model id invalid, yielding an Invalid model format error.

What the user should do

Update the model value from provider::model to provider:model (single colon). Configure clients to use the model ids shown in Cherry Studio's gateway/model list, which already use the single-colon form.

Notes for release manager

Parsing lives in src/main/features/apiGateway/services/ProxyStreamService.ts (indexOf(':')). A leading or trailing colon (":m" / "p:") is rejected with Invalid model format. Covered by the model-id parse tests in services/__tests__/ProxyStreamService.parse.test.ts.