receive: Add TSDB commit duration metric to receive writer

[yishi3-db]

Changes

Add a new Prometheus histogram metric thanos_receive_writer_commit_duration_seconds to track
the latency of TSDB Commit() calls in the receive writer.

Previously there was no way to observe how quickly metric samples are persisted to disk.
The existing thanos_receive_write_e2e_latency_seconds covers full end-to-end ingestion latency
but does not isolate the disk commit step. This metric makes that visible so we can track
changes in storage performance over time.

Verification

1. Tested with unit tests (go test -run TestWriter ./pkg/receive/...).
2. Passed CI
3. Verified in integtest env, saw metrics got emitted

[willh-db]

Cool!

[yuchen-db]

Seems to be a high cardinality metric?

[willh-db]

Seems to be a high cardinality metric?

@yuchen-db cardinality should be (16 histogram buckets) * (# of db pods) so thanos_receive_writer_commit_duration_seconds_bucket has an upper bound of ~15,000 time series in the biggest clusters unless I am missing something