Implement adaptive throttling phases 2-4 with quality monitoring

Phase 2: Transport-specific limits
- Add USB TMC, Serial, and TCP/IP specific queue depth and interval limits
- USB TMC: Conservative limits (5 queue depth, 1-2s intervals)
- Serial: Standard limits (10 queue depth, 0.5-1s intervals)
- TCP/IP: Higher buffering (15 queue depth, 0.5-1s intervals)
- Extract and pass transport_type through connection chain

Phase 3: Enhanced quality scoring
- Implement weighted recent history (2.0x for most recent 25%)
- Add quality trend detection using linear regression
- Quality tiers: excellent/good/fair/poor/critical
- Handle startup case with has_sufficient_data() check
- Override quality to critical for dead devices

Phase 4: Monitoring & metrics
- Extend DeviceMetrics with throttling metrics
- Add /metrics and /metrics/{device_id} API endpoints
- Record throttle events, backoff multiplier, quality scores
- Create comprehensive ADAPTIVE_THROTTLING.md documentation

Bug fixes:
- Fix empty dict reference bug in unified_scheduler (device_connections)
- Add missing time import in api.py
- Circuit breaker only blocks "dead" devices, not "unknown"
- Add recovery attempt logic (retry every 30s)
- Reduce circuit breaker log spam (200 skips vs 20)
- Fix frozen quality scores for dead devices
- Improve startup quality score handling

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

Author: MarkoVcode

benchmesh-serial-service/ADAPTIVE_THROTTLING.md

@@ -1,5 +1,12 @@
 version: 1
 devices:
+- id: eol-1
+  name: OWON OEL
+  driver: owon_oel
+  port: /dev/ttyOEL1515
+  baud: 115200
+  serial: 8N1
+  model: OEL1515
 - id: dmm-1
   name: OWON XDM
   driver: owon_xdm

benchmesh-serial-service/config.yaml

@@ -1,6 +1,7 @@
 import os
 import sys
 import json
+import time
 import asyncio
 import subprocess
 import hashlib
@@ -310,6 +311,50 @@ def list_drivers():
 
     return drivers
 
+@app.get("/metrics", summary="Get performance and throttling metrics for all devices", response_model=dict)
+def get_all_metrics():
+    """
+    Phase 4: Get performance and adaptive throttling metrics for all devices.
+
+    Returns metrics including:
+    - Serial port utilization %
+    - QPS (queries per second)
+    - API latency percentiles (P95, P99)
+    - Average queue depth
+    - Throttle events and skip rate
+    - Backoff multipliers
+    - Connection quality scores, tiers, and trends
+    - Transport types
+    """
+    global _manager
+    if not _manager or not _manager.metrics_collector:
+        return {"error": "Metrics collector not available"}
+
+    summary = _manager.metrics_collector.get_utilization_summary()
+    return {
+        "summary": summary,
+        "timestamp": time.time()
+    }
+
+@app.get("/metrics/{device_id}", summary="Get performance and throttling metrics for a specific device", response_model=dict)
+def get_device_metrics(device_id: str):
+    """
+    Phase 4: Get performance and adaptive throttling metrics for a specific device.
+
+    Returns detailed metrics for the specified device including utilization,
+    latency, queue depth, throttling events, backoff state, and quality metrics.
+    """
+    global _manager
+    if not _manager or not _manager.metrics_collector:
+        return {"error": "Metrics collector not available"}
+
+    metrics = _manager.metrics_collector.get_device_metrics(device_id)
+    if not metrics:
+        return {"error": f"No metrics available for device {device_id}"}
+
+    metrics["timestamp"] = time.time()
+    return metrics
+
 @app.get("/drivers/{driver_id}", summary="List models for a specific driver", response_model=list)
 def list_driver_models(driver_id: str):
     """