feat(gcpvertexai): do HTTP 429 like retries for Anthropic API HTTP 529 overloaded status code

[uddhav]

GCP Vertex AI: Add retry handling for Anthropic API 529 overloaded status code

Adds support for handling Anthropic's HTTP 529 "API Overloaded" status code in the GCP Vertex AI provider. This status code indicates temporary backend capacity issues rather than quota exhaustion.

Changes

• Added detection and retry logic for 529 "API overloaded" responses
• Applied the same backoff strategy used for 429 rate limit errors
• Enhanced error messages to distinguish between rate limits and overloaded states
• Added unit test to verify correct 529 status code handling
• chore: cleaned up the deprecated GCP Vertex AI model ID for Gemini 2.0 Pro Experimental

This improves reliability when interacting with Anthropic models through the Vertex AI provider during high-traffic periods.

[baxen]

We're doing a cleanup run on providers for #2953, since this is now a bit out of date with other changes i suggest we close this PR and include this in the refactor? Thank you for the contrib!