HDDS-16256. Avoid the per-group list copy in OMKeyRequest.filterOutBlocksStillInUse

[shuan1026]

What changes were proposed in this pull request?

OMKeyRequest.filterOutBlocksStillInUse called OmKeyLocationInfoGroup.createLocationList() on every version group of the reference key just to build a Map<ContainerBlockID, OmKeyLocationInfo> lookup table, allocating a throwaway flattened List per group that is immediately re-streamed and discarded.

This PR replaces the flatMap(createLocationList().stream()) chain with nested stream flattening over getLocationLists(), removing the per-group list allocation while preserving Collectors.toMap duplicate-key semantics.

Additionally, a throwaway (not committed) JUnit micro-benchmark measured per-lookup-map-construction thread allocation for both implementations (JDK21.0.12, 50,000-iteration warmup then 500,000 measured iterations per shape, output maps asserted equal, volatile sink to block JIT elimination):

Key Shape	Before	After	Saved
1 group x 1 block	240 ns / 950 B	107 ns / 704 B	246 B
1 group x 4 blocks	171 ns / 1072 B	169 ns / 872 B	200 B
1 group x 32 blocks	1149 ns / 4736 B	1014 ns / 4200 B	536 B
4 groups x 4 blocks	504 ns / 3648 B	404 ns / 2608 B	1040 B
16 groups x 8 blocks	6973 ns / 21088 B	6449 ns / 16688 B	4400 B

What is the link to the Apache JIRA

https://issues.apache.org/jira/browse/HDDS-16256

How was this patch tested?

• TestOMKeyCommitRequest#testValidateAndUpdateCacheOnOverwrite
• TestOMKeyCommitRequest#testValidateAndUpdateCacheOnOverwriteWithUncommittedBlocks

[sravani-revuri]

Overall LGTM.