feat: expand MiniMax provider registry

[octo-patch]

Reason: add target provider/model to existing provider registry.

• Register MiniMax-M3 and MiniMax-M2.7 with the configured model metadata.
• Expose global_en and cn_zh OpenAI-compatible and Anthropic-compatible endpoint choices in the provider manifest and LLM settings form.
• Route configured MiniMax Anthropic-compatible bases through the native Anthropic client while preserving the OpenAI-compatible path.
• Add focused registry and client-routing regression tests.

Checks:

• git diff --check
• ruff check backend/app/services/llm/client.py backend/tests/test_llm_provider_registry.py
• PYTHONPATH=. pytest tests/test_llm_provider_registry.py
• npm run build --prefix frontend

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 336d63b6d5

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Normalize MiniMax Anthropic URLs before client selection

When an admin enters the full MiniMax Anthropic endpoint ending in /v1/messages (or /v1), this exact comparison returns false even though AnthropicClient._normalize_base_url() already supports those forms. The model then falls through to OpenAICompatibleClient, which appends /chat/completions and sends the wrong protocol to a URL like .../anthropic/v1/messages/chat/completions; normalize the candidate URL the same way before matching, or accept the documented full endpoint variants.

Useful? React with 👍 / 👎.

[octo-patch]

Thanks for the review. I normalized the MiniMax Anthropic base URL before endpoint matching in backend/app/services/llm/client.py (_normalize_anthropic_base_url), so admin-entered full endpoints like .../anthropic/v1/messages and .../anthropic/v1 now route to AnthropicClient instead of falling through to the OpenAI-compatible client. I added parametrized routing tests in backend/tests/test_llm_provider_registry.py. I ran py_compile, ruff check (target files clean), and an isolated import of the registry/routing tests — all pass. The Drone CI step is failing on the git clone/git checkout step (fatal: unable to read tree) for both the original and the new commit, which looks like a runner/clone infrastructure issue rather than a build or test failure.

[octo-patch]

Aligned the MiniMax-M3 registry pricing with the target provider config (input 0.6 / output 2.4 / cache_read 0.12 USD per million tokens) and removed the unused pricing_tiers_usd_per_million_tokens field from the model spec and manifest. Ran the provider registry tests (8 passed), py_compile, and git diff --check.

[octo-patch]

Updated the PR branch with the latest main to resolve the merge conflict in the LLM settings UI, while keeping the MiniMax provider changes limited to the provider registry, registry tests, and LLM settings provider hints. I ran PYTHONPATH=backend /tmp/recvoNoEohPuXr-py311/bin/python -m pytest backend/tests/test_llm_provider_registry.py -q, py_compile, ruff check --select E9,F63,F7,F82 for the touched backend files, git diff --check, and npm run build in frontend.