perf: optimize coordinator HistoricalManagement duty runtime

[jtuglu1]

Description

High-level changes

• BalanceSegments - computes cluster shape (historicals, segments) once per run and reuses it for both the segment-move budget and the summary log, instead of calling getNumHistoricalsAndSegments() twice.
• MarkOvershadowedSegmentsAsUnused - builds the set of datasources that actually have overshadowed segments first, then skips timeline construction entirely for every other datasource, both when scanning servers and when checking zero-replica segments.
• MarkEternityTombstonesAsUnused - groups overshadowed segments by datasource up front, then checks each candidate tombstone only against its own datasource's group, instead of scanning the full overshadowed-segment list per candidate. Also swaps the timeline-based findNonOvershadowedObjectsInInterval scan for a direct per-segment isOvershadowed check.
• RunRules - caches ruleHandler.getRulesWithDefault(datasource) per datasource in a map built during the segment loop, and reuses that same map for the later broadcast-datasource check. Previously rules were looked up once per segment and then a second time per datasource during the broadcast scan.
• PrepareBalancerAndLoadQueues - collectUsedSegmentStats now iterates ImmutableDruidDataSource objects directly, using getTotalSizeOfSegments() and getSegments().size(), instead of streaming and summing over every segment in each datasource's timeline.
• SegmentReplicationStatus - the constructor no longer makes two full ImmutableMap.copyOf defensive copies. It holds the caller's map directly (safe because it's rebuilt fresh each cycle and not mutated after this point) and computes the total-replica-count map in one pass instead of a separate forEach. Also adds computeSegmentStats(), which does unavailable, under-replicated, and deep-storage-only counting in one pass over used segments instead of three.
• StrategicSegmentAssigner - allTiersInCluster is now computed once in the constructor and stored as a field, instead of being rebuilt with Sets.newHashSet(cluster.getTierNames()) on every replicateSegment/replicateSegmentPartially call.

I've kept the logging changes in in-case we deem some of them are debug-level worthy. Otherwise, please ignore those diffs during the review as they will be removed before merge. Initial results have shown a decrease in total HistoricalManagement duty runtime of about 35% on a cluster with 4M segments.

Release note

Speed-up coordinator HistoricalManagementDuties duty group

---

This PR has:

• been self-reviewed.
  • using the concurrency checklist (Remove this item if the PR doesn't have any relation to concurrency.)
• added documentation for new or modified features or behaviors.
• a release note entry in the PR description.
• added Javadocs for most classes and all non-trivial methods. Linked related entities via Javadoc links.
• added or updated version, license, or notice information in licenses.yaml
• added comments explaining the "why" and the intent of the code wherever would not be obvious for an unfamiliar reader.
• added unit tests or modified existing tests to cover new code paths, ensuring the threshold for code coverage is met.
• added integration tests.
• been tested in a test Druid cluster.

[FrankChen021]

I have reviewed the code for correctness, edge cases, concurrency, and integration risks; no issues found.

Reviewed 15 of 15 changed files.

---

This is an automated review by Codex GPT-5.5