Add MiniMax-M3 image and video input support

[octo-patch]

Reason: MiniMax-M3 declares image and video modalities, but the existing HTTP backend rejected media attachments.

This change:

• sends documented image and video content parts for MiniMax-M3 on both official regional endpoints;
• lets the configured MiniMax provider use Patina's existing OCR image pipeline;
• validates local attachment types and documented size limits while keeping media disabled for other models and endpoints; and
• covers local image encoding, video URLs, regional routing, and the non-target guard with focused tests.

Checks:

• npm run lint
• node --test tests/e2e/backends.test.js
• npm test (1,651 tests passed; the unrelated existing HOME/cwd deduplication case in tests/unit/config.test.js failed)

[vercel]

@octo-patch is attempting to deploy a commit to the devshwa's projects Team on Vercel.

A member of the Team first needs to authorize it.