request-logger: add Grok support

[gokhan]

Summary

Adds Grok (the xAI coding CLI) to the request-logger catalogue, in the same shape as Copilot: one provider, one env override, no config file.

• Grok builds the system prompt on the machine and sends it over HTTP to cli-chat-proxy.grok.com, so it can be logged. A grok.com login and an XAI_API_KEY both talk to that host; the API key is a credential fallback, not a different route.
• The printed command is GROK_CLI_CHAT_PROXY_BASE_URL=http://localhost:PORT/v1 grok. The /v1 suffix is required: Grok appends /responses (and /models) to the default https://cli-chat-proxy.grok.com/v1.
• The existing OpenAI renderer already reads the Responses API shape Grok uses by default.
• GROK_MODELS_BASE_URL is a different override (it switches the CLI onto API-key auth against a custom /v1/models endpoint). The command never sets it; a warning tells the student to unset it if it is already in the environment, or the logs folder stays empty.
• Grok also makes a second model call to title the session, so one turn can write two captures — the same fan-out as OpenCode. Documented in the README next to that note.

Test plan

• npx vitest run request-logger — 296 tests pass, including new Grok coverage: catalogue order, needsProvider, host, /v1 suffix, OpenAI renderer, POSIX and PowerShell commands, GROK_MODELS_BASE_URL not set, custom-target template reuse, customTargetNeedsModel false
• npx prettier --check on changed files
• Live grok -p turn through the proxy: POST /v1/responses returned 200, capture has tools + prompt + reply; GETs (/v1/models, /v1/settings) were forwarded and not logged