feat: Add metrics support to fs backend

[kfirtoledo]

Summary

Connect offloading connector metrics to fs_backend. Add monitoring.md with instructions for Grafana and Prometheus how to deploy.

Changes

• __init__.py: Update Python logger to use STORAGE_LOG_LEVEL for debugging
• worker.py: Populate TransferResult fields (size, time, type) for metrics export. Add per-request trace logging for transfer operations.
• C++ (logger.hpp, thread_pool.hpp): Move per-request debug prints to trace level to reduce log noise
• Monitoring:
  • Prometheus + Grafana stack with pre-configured KV offload dashboard
  • 9 panels with per-request/aggregate throughput, transfer counts, averages
  • docs/monitoring.md: Full deployment guide and KV cache offload benchmark

Metrics

Populate vLLM's built-in offloading metrics via TransferResult fields (size, time, type):

Metric	Type	Description
vllm:kv_offload_total_bytes	Counter	Total bytes transferred, labeled by transfer_type
vllm:kv_offload_total_time	Counter	Total time spent on transfers (seconds), labeled by transfer_type
vllm:kv_offload_size	Histogram	Distribution of transfer sizes in bytes, labeled by transfer_type

Grafana Dashboard

• Per-request throughput: increase(bytes) / increase(time)
• Aggregate throughput: rate(bytes) (30s default interval)
• Transfer count: increase(size_count)