Add Gemini Embeddings

[umaannamalai]

This PR adds instrumentation + tests for sync and async embeddings in google-genai.

[github-actions]

🦙 MegaLinter status: ✅ SUCCESS

Descriptor	Linter	Files	Fixed	Errors	Warnings	Elapsed time
✅ ACTION	actionlint	4		0	0	0.48s
✅ MARKDOWN	markdownlint	7	0	0	0	1.0s
✅ MARKDOWN	markdown-link-check	7		0	0	15.84s
✅ PYTHON	ruff	898	1	0	0	0.55s
✅ PYTHON	ruff-format	898	1	0	0	0.32s
✅ YAML	prettier	10	0	0	0	1.06s
✅ YAML	v8r	10		0	0	6.92s
✅ YAML	yamllint	10		0	0	0.52s

See detailed report in MegaLinter reports

[hmstepanek]

Main question was about the contents value and I had one suggestion about a refactor but totally don't have to do that in this PR.

[hmstepanek]

The if embedding_content else "" seems weird to me. Is this allowed to be a non string value? If it is None, then we are passing an empty string to the token count so we might as well not even call the token count function. Also it being None-is that option even allowed? How can you run an embedding without the input? Maybe this is just a leftover from before we added logic to not send value of type None and it's not needed anymore?

Suggested change

	embedding_content = str(embedding_content) if embedding_content else ""
	embedding_content = str(embedding_content) if embedding_content else ""

[umaannamalai]

Yeah, it seems that you can pass a single string or a list of strings (ex: https://github.com/googleapis/python-genai/blob/main/google/genai/models.py#L4028). My thought was to typecast to string in case it is a list so that we send the full input.

In terms of embedding_content being None, I was thinking more along the lines of there being a failure to grab the input from kwargs in the previous line:
embedding_content = kwargs.get("contents")
I was thinking down the road if they change the structure or argument names within kwargs then we may not get anything back if we try searching for contents within the request so this was to safeguard against that case.

[hmstepanek]

Ok that makes sense. In cases like that where the argument name might change, I think it's good to consider if it did-how would we catch it? Is this covering up that issue or exposing it so we can catch it quickly and fix it? Also, if this might be a list-how does that impact the token counting (might be something worth asking the AI team)? Would it be better to expand our spec for the token counting function to accept a list and cast it to a string after when recording the input into the event instead?

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 84.82143% with 17 lines in your changes missing coverage. Please review.

Please upload report for BASE (feature-gemini-instrumentation@2cadc57). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
newrelic/hooks/mlmodel_gemini.py	84.68%	12 Missing and 5 partials ⚠️

Additional details and impacted files

@@                        Coverage Diff                        @@
##             feature-gemini-instrumentation    #1345   +/-   ##
=================================================================
  Coverage                                  ?   81.46%           
=================================================================
  Files                                     ?      205           
  Lines                                     ?    22933           
  Branches                                  ?     3629           
=================================================================
  Hits                                      ?    18683           
  Misses                                    ?     3033           
  Partials                                  ?     1217           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[hmstepanek]

Had one minor fixup and some continued discussion on the contents parameter. If you want to merge this and then investigate that further with the AI team I think that's fine since it's going to a feature branch anyway.

[hmstepanek]

Ok that makes sense. In cases like that where the argument name might change, I think it's good to consider if it did-how would we catch it? Is this covering up that issue or exposing it so we can catch it quickly and fix it? Also, if this might be a list-how does that impact the token counting (might be something worth asking the AI team)? Would it be better to expand our spec for the token counting function to accept a list and cast it to a string after when recording the input into the event instead?