GCP: Route GCS batch deletes per credential prefix

[wombatu-kun]

Summary

Closes #16480

GCSFileIO.internalDeleteFiles partitioned BlobIds into fixed-size batches and then selected the GCS Storage client once per batch from only the first object's path. When a single GCSFileIO is configured with multiple per-prefix StorageCredentials (vended-credentials flow), a batch that crossed prefix boundaries was issued in full through whichever client matched the first object — sending the rest of the batch through the wrong credentials. The fix groups BlobIds by their PrefixedStorage client (via the existing longest-prefix-match clientForStoragePath helper) before partitioning into batches, and uses each client's own deleteBatchSize. No public API or exception contract changes.

This mirrors how S3FileIO.deleteFiles already groups by bucket before batching.

Tests

Two new unit tests in TestGCSFileIO:

• deleteFilesRoutesToCorrectClientPerPrefix — interleaves objects across two credential-prefixed buckets and asserts that each per-prefix Storage client receives only its own BlobIds.
• deleteFilesBatchesPerClient — sets a small gcs.delete.batch-size and asserts that batches stay per-client and never mix BlobIds from two prefixes.

Confirmed locally with ./gradlew :iceberg-gcp:test --tests "org.apache.iceberg.gcp.gcs.TestGCSFileIO" (26 tests, 0 failures). The new tests also fail when run against the previous code, confirming they catch the bug.

[wombatu-kun]

PR isn't the cause of Kafka Connect CI failure. I've fixed Kafka Connect integration test flakiness in separate PR #16438