feat: add minimax to the provider enum

[inference-gateway-maintainer]

Summary

Adds the minimax provider (introduced upstream in inference-gateway/schemas#62, closing inference-gateway/schemas#61) to the SDK. MiniMax is OpenAI-compatible: id minimax, bearer auth, base URL https://api.minimax.io/v1, supports_vision: true.

Changes:

• openapi.yaml (vendored spec): add minimax to the Provider enum, its x-provider-configs entry, and the MINIMAX_API_URL / MINIMAX_API_KEY config env vars — matching upstream byte-for-byte.
• inference_gateway/models.py: add minimax to the generated Provider enum (both Literal blocks), in spec order (after mistral, before moonshot).
• README.md: list MiniMax under "Supported Providers".

Scope note (focused sync)

The upstream schemas openapi.yaml has drifted beyond minimax since this copy was last vendored (header timestamp 2026-05-07). To keep this PR scoped to the issue, I synced only the minimax additions, not the full upstream delta. Deferred (unrelated) upstream changes, suitable for a separate "sync vendored openapi.yaml with schemas" task:

• streaming response switched from a bare $ref: SSEvent to oneOf: [SSEvent, CreateChatCompletionStreamResponse];
• content added to the streaming chunk delta's required (would change generated typing);
• moonshot supports_vision flipped false → true;
• trailing-whitespace tweaks in two descriptions.

Codegen note

datamodel-codegen is not runnable in the CI sandbox, so the Provider enum delta was applied directly to models.py. Because the vendored spec was updated in lockstep, the result is identical to generator output (enum order preserved) and task generate stays idempotent.

Validation

• models.py syntax-checked via ast.parse.
• The new enum members match the existing black formatting (indentation + trailing comma), so the CI checks (black --check ., pytest tests/) are expected to pass. Dev-dependency install and the codegen tool are blocked in this sandbox, so the full lint/test suite runs in CI.

Closes #42