HDDS-16281. Reduce per-key overhead in Recon reprocess tasks

[rich7420]

What changes were proposed in this pull request?

Recon full reprocess tasks repeatedly resolve worker-local accumulation maps through a shared concurrent map for every key. The OBS namespace summary task also performs RocksDB bucket-table point reads for every key, even when consecutive keys belong to the same bucket.

This change:

• caches OBS bucket metadata per worker and reuses the cached bucket object ID;
• uses ThreadLocal worker accumulation maps while retaining a concurrent registry for final merging and flushing;
• applies the worker-map optimization to namespace summary, container-key mapping, and file-size count reprocess paths;
• adds regression coverage for the OBS bucket lookup count.

What is the link to the Apache JIRA

https://issues.apache.org/jira/browse/HDDS-16281

How was this patch tested?

https://github.com/rich7420/ozone/actions/runs/32956525485

• 10 x 10 flaky-test-check: 100/100 passed after retrying one Maven Central HTTP 403 infrastructure failure.
• A local before/after task-level benchmark used 100,000 OBS keys across two buckets, two warmups, and five measured runs, repeated in reverse order. Median reprocess time decreased from 2098/2109 ms on upstream/master to 1598/1548 ms with this patch (24-27% lower), while throughput improved by 31-36%.

Generated-by: Codex (GPT-5)

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.