Coverage gaps in `opentelemetry-instrumentation-google-genai` instrumentations

Using this to track some gaps as I go about monkey patching various APIs as part of the openinference donation stuff..

Gaps:
A generic way for the user to provide attributes that make it into the span.. We use a ContextVar to do this for the GenerateContent method, and could do the same for the other APIs..

Embeddings:

• We could allow capturing various EmbedContentConfig in attributes the same way we do GenerateContentConfig in the generate_content instrumentation.
• Openinference captures the actual raw embeddings in their instrumentation -- that seems too big to put into an attribute..
• https://github.com/googleapis/python-genai/issues/2658 -- EmbedContentResponse has a bug that doesn't let us get tokens from usage_metadata

Interactions:

• Need to capture interaction ID: https://github.com/open-telemetry/semantic-conventions-genai/pull/372
• Expand input / output messages to capture all inputs / outputs -- the SDK accepts so many different types (related sem conv issue: https://github.com/open-telemetry/semantic-conventions-genai/issues/380)

The issue does not name repository files or tests. Start by locating the existing GenerateContent instrumentation and compare its ContextVar and config-attribute handling with the Embeddings and Interactions APIs. Review the linked semantic-conventions and SDK issues first; done requires agreeing on a focused scope and tests for the selected gaps.