perf(pruning): cache S3 directory listings by parent to reduce API calls

[khalid244]

Summary

• Optimized S3 directory existence checks by caching ListDirectories results at parent (day) level
• Previously each hour directory was checked individually, now all hours within a day share one API call

Results (30-day query)

• S3 API calls: 752 → 33 (23x reduction)
• Pruning time: ~38s → ~3s (12x faster)

Test plan

• Tested with 30-day query on production data
• Verified same pruning results (73 paths) before and after change

[xe-nvdk]

Hey @khalid244, thank you for the PR. I will review this PR.

[xe-nvdk]

Code Review: ✅ Approved

Excellent performance optimization! Thoroughly analyzed the implementation.

What this PR does

Caches S3 ListDirectories results at the parent (day) level instead of checking each hour directory individually. For a 30-day query:

• S3 API calls: 752 → 33 (23x reduction)
• Pruning time: ~38s → ~3s (12x faster)

Code Quality Assessment

Aspect	Status
Correctness	✅ Logic is sound
Error handling	✅ Safe fallback (assumes dir exists on error)
Cache key design	✅ Clean namespace (remote:parent:)
Thread safety	✅ Uses existing mutex-protected globCache
Tests	✅ All 28 pruning tests pass

Minor observations (not blockers)

1. On cache hit, rebuilds child set from slice - acceptable O(24) overhead
2. Old cache keys (remote:) expire naturally via TTL

Cost Impact

For S3-heavy workloads, this reduces LIST API costs significantly (~$0.005/1000 calls).

LGTM - merging.