fix: make gemini fail more less

[Abirdcfly]

What type of PR is this?

/kind flake

What this PR does / why we need it

1. Trigger a fast reconcile by returning an error when the state of the llm is not ready, instead of waiting 1 hour for a reconcile as if the state of the llm was ready.
2. Use different apikey in 3 tests at the same time, used to reduce the level of concurrency.
3. 3 retries after chat error

Which issue(s) this PR fixes

Fixes #

Special notes for your reviewer

In testing, there are still 2 relatively rare error cases (excluding errors returned by the LLM itself) that need to be fixed with subsequent pull requests

1. apiserver panic just log 2024-03-08T06:18:53.745667941Z stderr F 2024/03/08 06:18:53 googleapi: Error 503:
2. {"error":"run node chain-node: llmchain run error: Post \"https://generativelanguage.googleapis.com/v1/models/gemini-pro:generateContent?%24alt=json%3Benum-encoding%3Dint\u0026key=xxxxxxx\": context deadline exceeded"}