Add observability logs for load-based partition assignment

[dhananjay-sawner]

Summary

Adds INFO logs at the critical decision points of throughput-based partition assignment to verify that load balancing is actually working — i.e. distinguish true load-based (throughput-balanced) assignment from round-robin fallback, and observe whether the resulting per-task load is balanced.

Today the assigner logs min/max partitions across tasks, but load-based assignment intentionally makes partition counts uneven while balancing byte rate, so those counts do not prove balance. There is also no signal for whether throughput data was actually used, or whether an assignment fell back to round-robin.

Changes (logging only — no behavior change)

LoadBasedPartitionAssignmentStrategy

• Log whether an assignment is initial (throughput fetched → true load-based balancing) or incremental (throughput not refetched → newly unassigned partitions placed round-robin). Throughput is only fetched when no partitions are currently assigned (STOPPED → READY, first assignment, or restart).
• After a successful fetch, log throughput-provider coverage: provider class, cluster name, and number of partition entries returned.

LoadBasedPartitionAssigner

• Log throughput coverage: recognized partitions split into partition-level vs topic-level fallback, vs unrecognized (round-robin) partitions, plus provider entry count.
• Log the resulting per-task throughput distribution (min/max/avg/total/maxMinusMin in KBps) — the direct signal that byte-rate load is balanced across tasks.

Together these answer: which assignment mode ran → did the provider return data → how many partitions got real rates vs round-robin → is the resulting per-task load even.

Testing Done

• ./gradlew :datastream-server:compileJava :datastream-server:checkstyleMain pass.
• ./gradlew :datastream-server:test --tests "*TestLoadBasedPartitionAssigner" --tests "*TestLoadBasedPartitionAssignmentStrategy" — 15/15 pass.