[fix][broker] Fix bucket delayed message index metrics reset on scrape

[nodece]

Motivation

BucketDelayedMessageIndexStats used LongAdder.sumThenReset() for operation
counts and StatsBuckets.refresh() for latency buckets. Since both reset their
internal counters after reading, every stats collection cleared the accumulated
metrics. As a result, both topic stats and exported metrics only reported values
from the first read after new operations, while subsequent reads returned zero
until more operations occurred.

Modifications

• Add StatsBuckets.snapshot() to read cumulative values without resetting
  internal counters.
• Switch BucketDelayedMessageIndexStats to use cumulative reads for operation
  counts and latency buckets.
• Add tests covering repeated stats collection.