[PECOBLR-1146] Feature Flag Cache with Reference Counting

[samikshya-db]

Summary

Implement Phase 2 (Per-Host Management - Part 1) of telemetry for the Go driver: feature flag cache with reference counting to prevent rate limiting.

Changes

• Add featureFlagCache struct with singleton pattern for global cache management
• Implement featureFlagContext struct with per-host state (enabled flag, lastFetched, refCount)
• Add getOrCreateContext() to increment reference count when connections open
• Add releaseContext() to decrement reference count and cleanup when connections close
• Implement isTelemetryEnabled() with 15-minute TTL cache and automatic refresh
• Add fetchFeatureFlag() for HTTP integration with Databricks feature flag API
• Include fallback to cached value on fetch errors for resilience
• Add comprehensive unit tests (26 tests, 100% pass rate)

Key Features

• Per-host caching: Prevents rate limiting from multiple connections to same host
• Reference counting: Automatic cleanup when last connection closes
• 15-minute TTL: Automatic cache expiration and refresh
• Thread-safe: Uses sync.RWMutex for concurrent access
• Fail-safe: Falls back to cached value when HTTP fetch fails
• Non-blocking: Designed for lazy fetch when telemetry needs flag value

Testing

• ✅ All 26 unit tests passing (including 12 previous config/tags tests)
• ✅ Reference counting verified (increment/decrement/cleanup at zero)
• ✅ Cache expiration and refresh tested
• ✅ Concurrent access safety validated (100 goroutines)
• ✅ HTTP fetch success/failure scenarios covered
• ✅ Multiple host management tested
• ✅ Build successful with no errors

JIRA

Closes PECOBLR-1146
Part of PECOBLR-1143

Design Document

See telemetry/DESIGN.md Section 3.1 for detailed design specification.

🤖 Generated with Claude Code

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

[samikshya-db]

Closing because #304 is merged already