Python: Add Mistral AI embedding client package

[daric93]

Motivation and Context

Adds Mistral AI embedding support to the Python agent framework, as part of the Phase 9 additional embedding implementations tracked in #4171.

Mistral AI provides embedding models via their Python SDK. This PR adds a new agent-framework-mistral package with an embedding client that follows the same patterns established by the Ollama and OpenAI embedding clients.

Description

New package: python/packages/mistral/

• MistralEmbeddingClient — embedding client with OpenTelemetry telemetry support
• RawMistralEmbeddingClient — raw client without telemetry layer
• MistralEmbeddingOptions — TypedDict extending EmbeddingGenerationOptions for Mistral-specific options
• MistralEmbeddingSettings — TypedDict for env var resolution (MISTRAL_API_KEY, MISTRAL_EMBEDDING_MODEL, MISTRAL_SERVER_URL)

Uses the mistralai SDK v2 (>=2.0.0,<3) with create_async() for async embedding generation. Maps Mistral's output_dimension parameter to the framework's dimensions option, and normalizes usage fields to input_token_count / total_token_count.

Tests: 14 unit tests (100% coverage) + 1 integration test (skipped without API key).

CI: Added Mistral to the python-tests-misc-integration job in both python-integration-tests.yml and python-merge-tests.yml. Integration tests will skip until MISTRAL_API_KEY (secret) and MISTRAL_EMBEDDING_MODEL_ID (var) are provisioned in the GitHub integration environment.

Dependency note: mistralai pins opentelemetry-semantic-conventions<0.61, but the workspace dev dependency opentelemetry-sdk==1.40.0 requires 0.61b0. Resolved via override-dependencies in
pyproject.toml. Safe at runtime; removable once mistralai relaxes the bound.

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the Contribution Guidelines
• All unit tests pass, and I have added new tests where possible
• Is this a breaking change? No

[daric93]

@eavanvalkenburg please review when you have time — this implements the Mistral AI embedding client from #4171.