feat(collector): clean up expired tables during hotspot detection

[empiredan]

#2358

Currently, when collecting hotspot-related metrics, the Go collector only adds
new tables. If a large number of tables are deleted, the collector may still retain
those deleted tables in memory, leading to excessive memory usage.

This PR introduces a table cleanup mechanism: if no metric data has been collected
for a table over a prolonged period, the table will be removed to release the memory
it occupies.

[acelyc111-bot]

Review: Clean up expired tables during hotspot detection

Summary: Adds periodic cleanup of expired table analyzers in the hotspot partition detector, controlled by a new retention_period config (default 24h). Also fixes a TODO by using proper HostPort.Equal() comparison.

What's good:

• Clean implementation with proper goroutine lifecycle management (context + WaitGroup for graceful shutdown)
• retireExpiredTables uses a single lock acquisition per cleanup cycle — efficient
• Expiration timestamp is set at analysis time (not creation time), so active tables never expire
• Bonus: Fixed the TODO(wangdan) for HostPort.Equal() usage
• Config validation ensures RetentionPeriod > 0

Minor notes:

• The checkExpiration ticker interval equals RetentionPeriod (24h default) — this means tables could live up to 2x retention period in the worst case. Consider using a shorter check interval (e.g., RetentionPeriod / 4 or a fixed 1h) for tighter cleanup. Not a blocker though.
• Config file has some formatting cleanup (extra spaces removed) — fine, but ideally in a separate commit

Verdict: ✅ Approve — Well-structured, properly tested. Ready to merge.