[client] Implement adaptive fetch rate control for LogScanner

[swuferhong]

(The sections below can be removed for hotfixes or typos)
-->

Purpose

Linked issue: close #3006

For partitioned tables with many inactive partitions, the LogFetcher
previously sent fetch requests to all subscribed buckets equally,
wasting CPU and network resources on empty partitions.

Introduce BucketFetchRateController that uses exponential backoff to
reduce fetch frequency for buckets that consistently return no data.
Buckets that return data are always fetched at full frequency, and a
single non-empty fetch immediately resets the backoff.

Backoff schedule: 1, 2, 4, 8, 16, 32 rounds (configurable max).

New config options:

• client.scanner.log.adaptive-fetch.enabled (default: true)
• client.scanner.log.adaptive-fetch.max-skip-rounds (default: 32)

Brief change log

Tests

API and Format

Documentation