fix: add Primary-Wait Prometheus counters + conservative defaults — BIZ-60 review

P0 changes per 4-reviewer consensus (严维序/陆怀瑾/沈路明/梁思筑):

1. Prometheus metrics counters (proxy.py + server.py):
   - sidecar_primary_wait_enter_total: requests entering Primary-Wait
   - sidecar_primary_wait_recovery_total: successful primary recoveries
   - sidecar_primary_wait_exhausted_total: wait exhausted → emergency

2. Conservative default (config.py):
   - primary_wait_max_retries: 6 → 3 (15s total wait, safe start)
   - Observe recovery rate before increasing to 6

Counters form complete funnel: enter - recovery = exhausted,
enabling Grafana monitoring and ROI validation per COO/PM/Ops.

server.py

@@ -20,7 +20,13 @@ from crypto import init_crypto, is_initialized
 from pool_manager import PoolManager
 from rate_limiter import PerBackendRateLimiter
 from router import Router
-from proxy import handle_proxy_request, get_emergency_count
+from proxy import (
+    handle_proxy_request,
+    get_emergency_count,
+    get_primary_wait_enter_count,
+    get_primary_wait_recovery_count,
+    get_primary_wait_exhausted_count,
+)
 
 from storage.db import init_db, create_tables, run_integrity_check, get_connection, _DB_PATH
 from storage.backend_store import (
@@ -383,6 +389,11 @@ async def metrics() -> Response:
     # Emergency count (from proxy module)
     lines.append(f"sidecar_emergency_count {get_emergency_count()}")
 
+    # Primary-Wait metrics
+    lines.append(f"sidecar_primary_wait_enter_total {get_primary_wait_enter_count()}")
+    lines.append(f"sidecar_primary_wait_recovery_total {get_primary_wait_recovery_count()}")
+    lines.append(f"sidecar_primary_wait_exhausted_total {get_primary_wait_exhausted_count()}")
+
     # DB sizes
     from storage.db import get_db_sizes
     sizes = get_db_sizes()