Add rate-limited and aggregate logging to reduce log volume at high load

[dsingal0]

Summary

At high load, logging every request can be overwhelming for logging solutions like Loki. This PR introduces two new CLI arguments to provide flexible logging controls:

New CLI Arguments

• --log-sample-rate: Controls maximum request success logs per second (default: 10, set to 0 to disable per-request logging)
• --log-aggregate-interval: Interval in seconds for logging aggregate statistics (default: 0 = disabled)

Features

Rate-Limited Logger

• Limits per-request success logs to a maximum number per second
• Prevents log flooding during traffic spikes
• Backward compatible (defaults to 10 logs/sec)

Aggregate Logger

• Tracks request statistics and logs aggregated summaries at regular intervals
• Provides periodic summaries like:

  Request aggregate: 5234 requests (87.2/s), 12 errors | 10456789 chars (174280/s) | 2345678 tokens (39094/s)
  

• Includes request count, error count, characters processed, and tokens processed per interval

Usage Examples

# Disable per-request logging, use aggregate logging every 60 seconds
text-embeddings-router --model-id model --log-sample-rate 0 --log-aggregate-interval 60

# Hybrid: max 2 individual logs/sec + aggregate stats every 30 seconds
text-embeddings-router --model-id model --log-sample-rate 2 --log-aggregate-interval 30

Both options can also be configured via environment variables:

• LOG_SAMPLE_RATE (default: 10)
• LOG_AGGREGATE_INTERVAL (default: 0)

Files Changed

• router/src/logging.rs - Added RateLimitedLogger and AggregateLogger implementations
• router/src/main.rs - Added CLI arguments
• router/src/lib.rs - Updated run() signature to pass loggers
• router/src/http/server.rs - Updated all 6 endpoint handlers to use new loggers
• README.md - Added comprehensive documentation for the new features

Benefits

✅ Reduces logging overhead at high load
✅ Provides better observability through aggregate statistics
✅ Backward compatible - defaults similar to current behavior
✅ Works seamlessly with Loki and other log aggregation solutions
✅ Configurable via CLI args or environment variables