feat(llm_api): add MiniMax provider to synced LLM helper

[octo-patch]

Reason: provider-add — add a MiniMax provider to the synced LLM helper for the MiniMax-M3 / MiniMax-M2.7 models and their OpenAI-compatible endpoint.

Changes

Add a minimax provider to the shared llm_api.py tool starter and to the identical synced copies shipped with the built-in packs and the VS Code extension. The new provider follows the existing pattern used by the OpenAI-compatible providers (e.g. deepseek, siliconflow):

• create_llm_client("minimax") builds an OpenAI client pointed at the MiniMax global OpenAI-compatible endpoint (https://api.minimax.io/v1) using the MINIMAX_API_KEY environment variable.
• query_llm defaults the text model to MiniMax-M3 and routes minimax through the existing OpenAI-style chat completions path (which already supports text and image inputs).
• The CLI --provider choices now include minimax, with MiniMax-M3 as the default model.
• .env.example documents the new MINIMAX_API_KEY variable.

Files updated (all kept byte-identical, as in the existing repo layout):

• tools/llm_api.py
• rule_sets/tool_starters/llm_api.py
• src/rulebook_ai/packs/light-spec/tool_starters/llm_api.py
• src/rulebook_ai/packs/medium-spec/tool_starters/llm_api.py
• src/rulebook_ai/packs/heavy-spec/tool_starters/llm_api.py
• vscode-extension/manage_rules/python/tool_starters/llm_api.py
• vscode-extension/manage_rules/python/rule_sets/tool_starters/llm_api.py
• .env.example

Checks

• python3 -m py_compile on all seven llm_api.py copies — passed.
• Smoke test importing llm_api.py and exercising create_llm_client("minimax"): missing key raises ValueError; with a key it returns an OpenAI client with base_url=https://api.minimax.io/v1 — passed.
• Secret scan of the diff against the configured regex — clean.