Salvage of #12829 by @Tranquil-Flow onto current main.

Summary

hermes model MiniMax-M2.7 now succeeds. Previously, validation hard-rejected every MiniMax model because MiniMax's /v1/models endpoint returns 404 and commit aeb5313 flipped probe failure from warn-and-accept to hard-reject.

Changes

• hermes_cli/models.py: new minimax / minimax-cn branch in validate_requested_model(), placed before the live probe — mirrors the existing openai-codex pattern. Case-insensitive exact match, case-insensitive auto-correct at 0.9 cutoff, similar-model suggestions at 0.5, unknown names accepted with a warning (catalog may be incomplete).
• tests/test_minimax_model_validation.py: 7 tests covering exact match, case-insensitivity, near-match suggestion, unknown-model acceptance with warning, and catalog-path-is-taken guard.

Validation

• Targeted pytest: 70/70 passed across tests/test_minimax_model_validation.py + tests/hermes_cli/test_model_validation.py.
• E2E with fetch_api_models stubbed to return None: exact/lowercase/uppercase all accepted+recognized, MinMax-M2.7 auto-corrects to MiniMax-M2.7, unknown models accepted with the "MiniMax does not expose a /models endpoint" warning, minimax-cn behaves the same.

Closes #12611
Closes #12460
Closes #12399
Closes #12547

Supersedes #12564 (duplicate, same root cause, less coverage).