[fhgffy]

Description

Embedding requests fail before the API call because the AI SDK adapter supplies a language model to embed and embedMany. Select the provider's embedding model and use the configured model ID without its provider prefix.

Closes #13370.

Checklist

• I've read the contributing guide.
• No documentation change is needed; this repairs the existing embedding path.
• Added single-input and batch embedding regressions.

Tests

• The regressions use the real AI SDK OpenAI provider with local mock HTTP responses, and check the request path, model, inputs, response ordering, and token usage.
• Full packages/openai-adapters suite: 147 passed, 5 skipped.
• Complete TypeScript build with --preserveSymlinks for the local dependency junction: passed.