[octo-patch]

Reason: Add MiniMax text model configuration with MiniMax-M3.1-Flash as the simple-task default and MiniMax-M3 for complex tasks.

Purpose

• Make current MiniMax text models available through RAI configuration while preserving saved and explicitly requested model IDs.

Proposed Changes

• Add MiniMax initialization with configurable chat protocol and regional endpoints, using MINIMAX_API_KEY unless the caller supplies a key.
• Set repository and generated configuration defaults to MiniMax-M3.1-Flash and MiniMax-M3. Existing configuration files remain unchanged unless explicitly regenerated with --force.
• Keep embeddings assigned to a separate supported vendor. Document configuration-file setup; the optional configurator does not currently list MiniMax.
• Add offline tests that capture serialized HTTP requests from the real SDK clients: defaults, both protocols and configured regions, saved/custom model names and endpoints, explicit API keys, reasoning options, tool history, and image payloads.
• Include the required dependency lock and minor version update. Merge current main without rewriting published history.

Issues

• None.

Testing

• Passed: uv lock --check --offline (365 packages resolved).
• Passed: git diff --check, credential-pattern scan, and repository/generated MiniMax configuration consistency check.
• Passed: preliminary lint with the repository rule selection using the available Ruff 0.16.6; final Ruff 0.9.4 validation is pending.
• Pending: focused initialization tests and pinned formatter/linter; the isolated lightweight dependency setup is waiting for the shared installation lock.
• Not run: full uv sync --all-groups, ROS build, and full suite. ROS 2 Jazzy is absent locally, and heavy setup is paused by the workspace disk-capacity gate.
• No paid inference, live regional API verification, physical robot execution, or tool execution. Request tests intercept the HTTP client before any network request.