[Feature] Document a custom Tsubasa chat-completions provider

[cenab]

Area

AI providers and models; Documentation

Problem

Tsubasa users need a provider configuration with a dedicated key, public aliases and a 32,768-token context window. Prime Agent already provides the needed transport through custom models.json; there is no missing protocol.

Proposed direction

Would a small example in the existing custom-model guide be useful once service qualification is available? It would declare providers.tsubasa with api="openai-completions", baseUrl="https://api.tsubasa.sh/v1", apiKey="TSUBASA_API_KEY", and the two aliases. Model tool/reasoning capabilities would not be inferred from the wire protocol.

If invited, test the actual ModelRegistry loader, both model IDs, bearer selection, normal/streaming requests and the agent prompt budget before a PR. No native provider implementation or dependency would be added.

Alternatives considered

Keep the JSON solely in the provider's public configuration catalog. A new native transport wrapper would duplicate the existing one.

Additional context

The tested public Tsubasa /v1/models route returned HTTP 404; there is no live inference proof. This Discussion requests direction, not an exception to the contribution/vouch gate. No maintainer invitation or vouch is claimed, and no unsolicited PR has been opened.

AI disclosure: OpenAI Codex inspected the source and drafted this Discussion.

Related generic documentation proposal: discussion #2020. This request is limited to a named service profile and its model-budget metadata.