fix(sidecar-v2): second-round review fixes

- cooldown_manager: move function-level imports to module top
- proxy.py: emergency_count counter now actually increments
- server.py: metrics reads emergency_count from proxy module
- dashboard.html: real JS CDN fallback (not just comment)
- requirements.txt: remove unused prometheus_client

Round 2 review residual fixes from 沈路明/陆怀瑾/梁思筑 feedback

Co-authored-by: multica-agent <github@multica.ai>

services/nvidia_sidecar/cooldown_manager.py

@@ -4,7 +4,7 @@ import time
 from datetime import datetime, timezone
 import structlog
 from config import config
-from storage.backend_store import set_backend_cooldown, clear_backend_cooldown
+from storage.backend_store import set_backend_cooldown, clear_backend_cooldown, get_backend
 from storage.cooldown_store import log_cooldown_event, end_cooldown_event
 
 logger = structlog.get_logger("sidecar_v2.cooldown_manager")
@@ -57,7 +57,6 @@ def check_and_clear_cooldown(backend_id: str) -> bool:
 
     Returns True if cooldown was cleared (backend is back online).
     """
-    from storage.backend_store import get_backend
     backend = get_backend(backend_id, decrypt_key=False)
     if backend is None:
         return False
@@ -97,7 +96,6 @@ def precheck_cooldown(backend_id: str) -> bool:
     If cooldown will expire within config.cooldown_precheck_threshold_seconds,
     skip the backend so we don't hit it again right as it expires.
     """
-    from storage.backend_store import get_backend
     backend = get_backend(backend_id, decrypt_key=False)
     if backend is None or backend.status != "cooling":
         return False