[codex] Document OpenAI-compatible provider setup

[Lee-take]

Summary

• Adds a clearer custom OpenAI-compatible provider example using CLI flags and environment variables.
• Documents persistent provider-scoped configuration for base URL, model, and API key.
• Lists common pitfalls around provider selection, model routing, and secret handling.

Fixes #1491

Validation

• git diff --check

[gemini-code-assist]

Code Review

This pull request updates the documentation in docs/CONFIGURATION.md by adding CLI and TOML configuration examples for OpenAI-compatible providers. It also introduces a section on common mistakes, covering topics such as provider naming, legacy base URL settings, and proper endpoint formatting. I have no feedback to provide as there were no review comments.

[Hmbown]

Independent review:

Verified the +37 line diff in docs/CONFIGURATION.md — adds a deepseek auth set CLI snippet, a second TOML block showing [providers.openai].model precedence, a TOML table-order tip, and a "common mistakes" list.

• Substantively accurate at the time of authoring; the technical advice (don't invent custom provider names, put base_url under [providers.openai], [providers.openai].model overrides default_text_model) is correct.
• v0.8.48 (pr-2256) already documents the same ground in docs/CONFIGURATION.md at lines ~93-115: same TOML template, same "do not invent a custom provider name", same model-override note.
• The added CLI invocations (deepseek auth set, deepseek --provider openai) are now stale — v0.8.48 renamed the binary to codewhale and prefers CODEWHALE_BASE_URL (with OPENAI_BASE_URL/DEEPSEEK_BASE_URL kept as legacy aliases).
• The "common mistakes" bullet list is the only piece with marginal incremental value over v0.8.48's prose, but the base_url should normally end at /v1 tip is already implicit in the existing example.
• Merge sim against main: clean (single-file doc merge).

v0.8.48 (#2256) compatibility: superseded — same guidance lives in v0.8.48's docs/CONFIGURATION.md, and 1867's CLI snippets use the pre-rename deepseek binary. Recommend close-as-superseded; if you want to keep the "common mistakes" framing, port that bullet list onto current codewhale-named commands.

[Hmbown]

@Lee-take — appreciate the doc contribution. The OpenAI-compatible provider setup guidance you wrote is now in v0.8.48's docs/CONFIGURATION.md (lines ~93–115) with equivalent content. Specifically the section that walks through provider = "openai" + custom base_url + model setup.

One note — the CLI snippets in your diff still use the pre-rebrand deepseek binary name; the canonical command is now codewhale (with deepseek shipping as a deprecation shim through v0.9.0). If you'd like to refresh the snippets and re-target the contribution at any gap you see in the current docs (issue #1491 specifically asks for a third-party tutorial), that'd be welcome. Otherwise this can be considered done.

Thanks for caring about the docs surface — it's perpetually undervalued.

[Lee-take]

Thank you for the careful review and for pointing to the current v0.8.48 docs.

I agree this PR is now superseded: the OpenAI-compatible provider setup is covered in the current docs/CONFIGURATION.md, and the CLI snippets in this diff still use the pre-rebrand deepseek command rather than the canonical codewhale command.

I'll close this PR as superseded instead of continuing to rework the old doc diff. If I revisit this area, I'll do it as a fresh contribution against current main, using codewhale commands and targeting a real remaining docs gap such as the third-party provider tutorial requested in #1491.

Appreciate the clear guidance here, and the note about docs being undervalued.