feat: added Google GenAI client; simplified IAI/clients API surface.

[bwplotka]

📑 Description

Disclaimer: I am not a GenAI experts, or official Google delegate around those topics. I know what you know from public docs (: I am simply contributing some improvements here as I experiment with the OSS tooling around analyzing Kubernetes problems. Thought I could improve some CNCF software on the way.

This PR adds the newest https://ai.google.dev/ API (for Gemini and PaLM (legacy) models).

I think it might be fixing #799, since the Gemini API added here is also supported by Google Cloud Vertex AI but there is no Vertex AI client/API per se (AFAIK), there was only PaLM APIs, so not sure. Sorry @swastik959 if I accidentally solved the issue you were assigned to. I had to do this work anyway for my experiments.

On the way I couldn't understand why IAI interface is so complex. After investigation it seems it might be some tech debt from initial experiments, I proposed some changes to limit code duplication, improve readability and consistency (and fixed some gaps on the way). Specifically:

• add auth flags specify which backend cares about which settings
• AI clients are not expected to parse multiple "failure texts", process prompt template and cache. This is common behavior that might be better to be implemented once on Analysis function. This fixed numerous existing bugs with some providers/clients which had inconsistent logic for that (lack of caching/lack of processing/caching is not best effort etc). This was offloaded to one getAIResultForSanitizedFailures function. Thanks of that GenerateComplete is trvial and Parse can be removed.
• AI clients are not aware of language. Current code only tries to force language with prompt prefix, so no need to inject it (we can inject later if needed). This simplifies Configure method.

✅ Checks

• My pull request adheres to the code style of this project
• My code requires changes to the documentation
• I have updated the documentation as required
• All the tests have passed

ℹ Additional Information

Gemini API is still rolling out e.g. it's not available in UK yet 🙈 Full details: https://ai.google.dev/available_regions

[codecov]

Codecov Report

Attention: 42 lines in your changes are missing coverage. Please review.

Comparison is base (526e22f) 46.54% compared to head (8737cc4) 45.87%.

Files	Patch %	Lines
pkg/analysis/analysis.go	0.00%	40 Missing ⚠️
pkg/server/analyze.go	0.00%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #829      +/-   ##
==========================================
- Coverage   46.54%   45.87%   -0.67%     
==========================================
  Files          27       27              
  Lines        2198     2230      +32     
==========================================
  Hits         1023     1023              
- Misses       1137     1169      +32     
  Partials       38       38              

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

[arbreezy]

Great work thanks @bwplotka.

Lgtm although I can't test the new backend.
One minor thing, our semantic PR validation fails if you have a capital letter after the prefix.

[bwplotka]

Done, thanks!

[matthisholleville]

Thank you @bwplotka for your contribution ! Good job ! Shouldn't we add an example in the readme? https://github.com/k8sgpt-ai/k8sgpt/blob/main/README.md#key-features

[bwplotka]

Yup, I can do that in later PR or this one, up to you. Tomorrow though, closing laptop for today!

[thschue]

lgtm

[thschue]

Yup, I can do that in later PR or this one, up to you. Tomorrow though, closing laptop for today!

Please open a new PR for the docs

Thank you for this contribution!

[bwplotka]

Thanks! Docs in progress 💪🏽