Reorganize layer cache Grafana dashboard for improved usability

Complete overhaul of the layer cache metrics dashboard to improve
readability, add missing operational metrics, and organize panels by
workflow.

Template Variables:
- Add operation template for filtering persistence latency metrics
- Convert backend template from hardcoded to dynamic query discovering
  all backends (foyer, postgres, s3) automatically
- Expand service template to include read-only services (edda) using
  union query across multiple metrics

Panel Organization:
Restructured 42 panels into 5 logical sections matching operational
workflows:

1. OVERVIEW & HEALTH - System health scorecard with error rates, queue
   depths, resolution rates, and write throughput balance
2. READ PATH PERFORMANCE - Request latency, backend latency comparison,
   cache miss rates, fallback tracking
3. WRITE PATH - S3 PERSISTENCE - Queue depth, backoff state, write
   throughput, error breakdown, dead letter queue
4. WRITE PATH - POSTGRES PERSISTENCE - Persister operations,
   persistence latency, retry queue depth and rates
5. EVICTION OPERATIONS - Eviction rates, failures, latency, and
   memory-only evictions

Readability Improvements:
- All p50/p95/p99 panels now show exactly 3 data series using panel
  repeat and aggregation
- Panel 26 (Persistence Latency): Repeats by operation, aggregates
  other dimensions
- Panel 27 (Backend Read Latency): Aggregates across caches
- Panel 28 (Request Latency): Repeats by cache
- Panel 29 (Backend Hit Latency): Repeats by cache, shows all backends
  together, removed miss series

New Metrics (8 panels):
- System Health Scorecard: Overall error rate, S3 queue depth, retry
  queue depth with color-coded thresholds
- Fallback Tracking: S3→PostgreSQL fallback rate indicating data
  availability issues
- S3 Write Error Breakdown: Stacked timeseries by error category
- S3 Dead Letter Queue Depth: Corrupted write tracking
- Write Throughput Balance: Compares S3 vs PostgreSQL write rates for
  dual-write monitoring
- Memory-Only Evictions: Tracks Foyer memory evictions without disk
  persistence
- Overall Cache Miss Rate: Single aggregated panel showing complete
  misses per cache

Removed Redundant Panels (3):
- Retry Queue Success vs Failure Ratio (data visible in other panels)
- S3 Write Failures by Error Category (replaced by comprehensive
  breakdown)
- Persister Retry Operations (superseded by specific retry queue
  panels)

Technical Details:
- Overall cache miss uses layer_cache_request_latency_ms metric
  (verified in layer_cache.rs:347-351) tracking complete misses across
  all layers
- All backends explicitly report both hit and miss metrics (verified in
  layer_cache.rs:154-308)
- Template union queries ensure all backends and services appear
  dynamically

Author: jhelwig