HDDS-16202. Avoid the list copy in ContainerEndpoint.getBlocks

[shuan1026]

What changes were proposed in this pull request?

ContainerEndpoint.getBlocks called OmKeyLocationInfoGroup.getLocationList() and then stream().filter().collect(toList()) on the result, allocating two throwaway lists per key just to walk each block once. getLocationList() is documented as expensive: it flattens locationVersionMap.values() into a brand-new List on every call, instead of the O(1) getLocationLists() accessor.

This PR rewrites getBlocks to iterate getLocationLists() directly with a nested loop and an inline containerID check, removing both list allocations. Same pattern as OMKeyRequest.sumBlockLengths (HDDS-16183)
and KeyManagerImpl.refreshPipeline (HDDS-5384). Output ordering and contents are unchanged.

What is the link to the Apache JIRA

https://issues.apache.org/jira/browse/HDDS-16202

How was this patch tested?

Existing tests cover getBlocks through the Recon GET /api/v1/containers/:id/keys endpoint, including block-id and pagination assertions.

The following test suite passed (2 tests):

• TestContainerEndpoint#testGetKeysForContainer
• TestContainerEndpoint#testGetKeysForContainerWithPrevKey

Also ran checkstyle.sh and author.sh successfully.

[sravani-revuri]

Overall LGTM.
Now getBlocks uses getLocationLists() instead of flattening and applies the container ID check during iteration.

[sodonnel]

LGTM - thanks for improving this.