Add cache source tracking (L1/L2) for Telegram bot responses

Implements cache tier identification in Telegram bot to display data source:
- "db" for database queries
- "cached L1" for in-memory cache hits
- "cached L2" for SQLite cache hits

Backend changes:
- Added cache metadata fields to TrendsResponse and DashboardSummary models
  (cache_hit, response_time_ms, cache_source)
- Updated /api/dashboard/summary and /api/dashboard/trends endpoints to
  include cache metadata when X-Include-Cache-Metadata header is present
- Cache metadata is extracted from request.state (set by @cached decorator)

Telegram bot changes:
- Updated API client to send X-Include-Cache-Metadata header
- Modified helpers to extract cache_source from backend responses
- Updated handlers to pass cache metadata to formatters
- Performance footer now displays specific cache tier (L1 vs L2)

Fixed Pydantic serialization issue:
- Changed field names from _cache_hit to cache_hit (without underscore)
- Pydantic excludes underscore-prefixed fields from JSON by default

🤖 Generated with Claude Code

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

reports-app/backend/app/models/dashboard.py

@@ -51,6 +51,11 @@ class TrendsResponse(BaseModel):
     metadata: Dict[str, Any]
     growth_rates: Optional[Dict[str, float]] = None
 
+    # Cache metadata (optional, for Telegram Bot)
+    cache_hit: Optional[bool] = None
+    response_time_ms: Optional[float] = None
+    cache_source: Optional[str] = None
+
 class DashboardSummary(BaseModel):
     """Model pentru toate datele dashboard-ului"""
     # CLIENȚI - statistici existente