WIP: Support GitHub's Responses Endpoint

[vtomnet]

This enables access to gpt-5-codex, with the following codex-cli config.toml. It works but needs polishing. TODO:

• isNonStreaming
• max_tokens
• type definitions
• testing

Config:

model_provider = "copilot-api"
model = "gpt-5-codex"
model_reasoning_effort = "high"

[model_providers.copilot-api]
name = "GitHub Copilot API"
base_url = "http://localhost:4141/v1"
wire_api = "responses"

Re: max_tokens, I'm not sure of the purpose of this code from chat-completions/handler.ts:

  if (isNullish(payload.max_tokens)) {
    const selectedModel = state.models?.data.find(
      (model) => model.id === payload.model,
    )

    payload = {
      ...payload,
      max_tokens: selectedModel?.capabilities.limits.max_output_tokens,
    }
    consola.debug("Set max_tokens to:", JSON.stringify(payload.max_tokens))
  }

Re: testing, the model is currently overloaded via GitHub so it's mostly unusable at the moment:

⚠️ stream error: We're currently experiencing high demand, which may cause temporary errors.; retrying 1/5 in 182ms…
⚠️ stream error: stream disconnected before completion: Transport error: error decoding response body; retrying 1/5 in 216ms…
⚠️ stream error: exceeded retry limit, last status: 429 Too Many Requests; retrying 1/5 in 214ms…

Another thing: Codex is unable to do web searches (--search), but it looks like a github limitation:

 ERROR  HTTP error: { error:
   { message: 'The requested tool web_search is not supported.',
     code: 'unsupported_value',
     param: 'tools',
     type: 'invalid_request_error' } }