Instrument `pydantic_ai.embeddings.Embedder` with `gen_ai.embeddings` spans

`Embedder.embed()`, `embed_query()`, and `embed_documents()` (and their sync variants) make model calls with no spans today.

OTel defines `gen_ai.embeddings` as a valid `gen_ai.operation.name` and we already have `OP.GEN_AI_EMBEDDINGS` defined in `sentry_sdk/consts.py`. Should create `gen_ai.embeddings` spans with model name, system, and token usage attributes.

Start with pydantic_ai.embeddings.Embedder.embed(), embed_query(), embed_documents(), and their sync variants, then inspect OP.GEN_AI_EMBEDDINGS in sentry_sdk/consts.py. Trace how model calls are instrumented and add spans with the requested model name, system, and token usage attributes; done means all listed methods emit gen_ai.embeddings spans.

由索引模型根據 Issue 內容生成。