[P1][Logs] Aggregate region anomaly logs instead of logging per object

Parent issue: #4691

Intent

Region-related anomalies such as slow regions, holes, stale requests, and untracked regions are currently prone to per-object logging. Under large region counts, the log volume grows with the number of affected regions instead of with the number of distinct incidents. This task moves those logs to aggregated summaries with small samples.

Modification points

• Replace per-region anomaly logs with window-based summaries plus top-N samples.
• Keep aggregate fields such as count, maxAge, largestGap, sampleRegionIDs, and sampleSubscriptionIDs.
• Reserve full per-region detail for debug or diagnostics-only paths.

Entry points

• logservice/logpuller/subscription_client.go:1055
• logservice/logpuller/subscription_client.go:1061
• logservice/logpuller/subscription_client.go:1065
• logservice/logpuller/subscription_client.go:1070
• logservice/logpuller/region_request_worker.go:280
• logservice/logpuller/region_request_worker.go:334
• logservice/logpuller/region_req_cache.go:166
• logservice/logpuller/region_req_cache.go:247

Done criteria

• Region anomaly logs no longer scale linearly with the number of affected regions on the default log path.
• The summary logs still expose the size, severity, and representative samples of the incident.
• Full detail is available only when explicitly requested through a debug or diagnostic path.

Start with the listed entry points in subscription_client.go, region_request_worker.go, and region_req_cache.go, tracing how region anomalies are currently logged. Check how the affected regions and subscriptions are collected, then verify that default logs use window summaries with counts, severity fields, and top-N samples while full detail remains limited to debug or diagnostic paths.