fix(bigtable): preserve resource labels on batcher metrics

[AliAbouelazm]

Fixes #18546.

The mutations batcher creates operation metrics directly through the metrics controller, bypassing the resource labels supplied by the table or authorized view. The missing project label causes the exporter to drop these metrics and log a warning on each export.

Create each batch's metric through the target's existing _create_operation helper. Metrics still start before waiting for flow control, so operation and throttling latency measurements keep their existing timing. Regenerate the synchronous implementation and tests with CrossSync.

Regression coverage includes tables and authorized views, default and explicit app profiles, multiple batches, and flow-control timing in both synchronous and asynchronous clients. The label regression fails before the fix in all eight configurations.

Validation:

• Full package unit tests on Python 3.12 with upb: 5,506 passed, 8 skipped.
• Full package unit tests on Python 3.14 with both Python and upb protobuf implementations: 5,506 passed, 8 skipped per run.
• Python 3.14 mypy: all 46 source files pass.
• Ruff import and format checks, Flake8, and package metadata checks pass.
• Independent offline validation confirms that the real mutation and export pipeline produces correctly labeled operation, attempt, and throttling metrics. All service calls are mocked.

The existing aggregate coverage gate remains below 99% on both untouched main and this branch: 98.330% before and 98.345% after. Every added executable line is covered.

Live Bigtable system tests were not run.

[gemini-code-assist]

Code Review

This pull request refactors the mutations batcher (both async and sync autogenerated versions) to pass the target object instead of the metrics_controller to add_to_flow_with_metrics. This allows the batcher to delegate operation creation directly to the target via _create_operation, enabling better tracking of resource labels (such as project, instance, table, and app profile IDs) on client-side metrics. Additionally, unit tests have been added to verify metric resource labels and flow throttling behavior. There are no review comments to address, and I have no further feedback to provide.