feat(sidecar-v2): implement multi-pool provider proxy with cooldown, rate limiting, WebUI

BIZ-52 Step3 开发实现:
- storage: backend/usage/cooldown/config CRUD with SQLite WAL
- crypto: AES-256-GCM API key encryption
- pool_manager: primary/fallback pool routing
- cooldown_manager: 429 exponential backoff cooldown
- rate_limiter: per-backend token bucket RPM control
- router: model → backend routing with pool priority
- proxy: multi-pool request forwarding with retry
- server: FastAPI admin API + OpenAI-compatible proxy + SSE
- dashboard: WebUI with provider CRUD, stats, charts

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

services/nvidia_sidecar/pool_manager.py

@@ -0,0 +1,83 @@
+"""Provider pool management: primary / fallback pool routing."""
+
+import structlog
+from typing import Optional
+
+from storage.backend_store import list_backends, get_pool_stats
+from storage.models import Backend
+
+logger = structlog.get_logger("sidecar_v2.pool_manager")
+
+
+class PoolManager:
+    """Manages provider pools and selects healthy backends for a given model.
+
+    Priority: primary pool → fallback pool.
+    Within a pool: healthy backends only, sorted by availability.
+    """
+
+    def __init__(self):
+        self._pool_order = ["primary", "fallback"]
+
+    def get_available_backends(
+        self, canonical_model: str, pool: Optional[str] = None
+    ) -> list[Backend]:
+        """Get all healthy, enabled backends that serve a model, in pool order.
+
+        Args:
+            canonical_model: Canonical model name to match.
+            pool: Optional pool filter (primary/fallback). None = all pools.
+
+        Returns:
+            List of ready backends sorted by pool priority, then RPM utilization.
+        """
+        backends: list[Backend] = []
+
+        pools_to_check = [pool] if pool else self._pool_order
+        for p in pools_to_check:
+            pool_backends = list_backends(pool=p, enabled_only=True, decrypt_key=True)
+            for b in pool_backends:
+                if b.status == "healthy" and b.has_model(canonical_model):
+                    backends.append(b)
+            if pool:
+                break
+
+        return backends
+
+    def get_any_healthy_backends(self, pool: Optional[str] = None) -> list[Backend]:
+        """Get all healthy, enabled backends regardless of model."""
+        backends: list[Backend] = []
+        pools_to_check = [pool] if pool else self._pool_order
+        for p in pools_to_check:
+            pool_backends = list_backends(pool=p, enabled_only=True, decrypt_key=True)
+            for b in pool_backends:
+                if b.status == "healthy":
+                    backends.append(b)
+            if pool:
+                break
+        return backends
+
+    def get_pool_status(self) -> dict:
+        """Get pool summary for dashboard."""
+        stats = get_pool_stats()
+        result = {}
+        for pool in self._pool_order:
+            s = stats.get(pool, {"total": 0, "enabled": 0, "healthy": 0, "cooling": 0, "error": 0})
+            result[pool] = s
+        # Also include any other pools
+        for pool, s in stats.items():
+            if pool not in result:
+                result[pool] = s
+        return result
+
+    def is_pool_available(self, canonical_model: str, pool: str = "primary") -> bool:
+        """Check if a pool has any healthy backends for a model."""
+        backends = self.get_available_backends(canonical_model, pool=pool)
+        return len(backends) > 0
+
+    def is_any_pool_available(self, canonical_model: str) -> bool:
+        """Check if any pool has healthy backends for a model."""
+        for pool in self._pool_order:
+            if self.is_pool_available(canonical_model, pool):
+                return True
+        return False