fix: Google Vertex embedMany sends oversized text-embedding batches (#19960)

## Background

Google Vertex embedding consumers could receive HTTP 400 responses
because embedMany sent 251 texts in a single :predict request instead of
respecting Vertex's 250-text limit.

## Root Cause

GoogleVertexEmbeddingModel hardcoded maxEmbeddingsPerCall as 2048 for
:predict models. The reproduction confirmed this caused one 251-item
request, while focused tests confirmed :embedContent models correctly
require single-item calls.

## Summary

Changed the advertised :predict embedding limit to 250, retained the
:embedContent limit of 1, removed reproduction-only artifacts, and added
a patch changeset.

## Testing

Added regression coverage for the 250-value :predict limit, both
single-value :embedContent model variants, and rejection at 251 values.

## End-to-end Validation

- `replay_original_reproduction` (`pnpm -C examples/ai-functions exec
tsx src/reproduction/google-vertex-embedding-batch-limit.ts`): exited
successfully without the original oversized-request signal.
- Live Vertex embedMany command: blocked before making a request because
`GOOGLE_VERTEX_LOCATION` was unavailable; no successful live validation
is claimed.

## Related Issues

Fixes #19952

Closes #19957

---------

Co-authored-by: ai-sdk-factory <308175966+ai-sdk-factory@users.noreply.github.com>
Co-authored-by: n1ckoates <58091943+n1ckoates@users.noreply.github.com>