Fix memory leaks and race conditions causing OOM shutdowns

[TheMarstonConnell]

Summary

• Queue race conditions: Added sync.Mutex to protect messages, stopped, and running fields accessed from multiple goroutines without synchronization
• Infinite retry loop: popAndPost() retried forever if blockchain was unreachable — capped at 10 retries, then releases waiting goroutines with error
• Unbounded queue growth: Post() now rejects new messages when queue exceeds 500, providing backpressure instead of growing until OOM
• Goroutine leaks from *gin.Context: ProcessFile, PostFile, and UpdateCollection no longer receive *gin.Context in background goroutines — dependencies are extracted before goroutine launch
• http.DefaultClient mutation: Replaced with package-level clients (uploadClient, cloneClient) with 120s timeouts
• Clone handlers with no timeout: http.Get() calls replaced with context-aware requests using cloneClient
• No DB connection pool limits: Set MaxOpenConns(25), MaxIdleConns(5), ConnMaxLifetime(5min)
• Storage purchaser silent exit: Changed return to continue on query error + added COALESCE for empty tables
• GetUsage returning nil, nil: Now returns proper error when no customer found

Test plan

• go build ./... passes (verified)
• go vet ./... passes (verified)
• Deploy to staging and monitor memory usage over 24h — should no longer see gradual memory growth
• Test upload flow end-to-end (single file, multi-file, clone from URL, IPFS clone)
• Test collection operations (add/remove files, nested collections)
• Verify queue backpressure: when queue is full, API returns error instead of hanging
• Simulate blockchain downtime: confirm queue stops retrying after 10 attempts instead of looping forever

🤖 Generated with Claude Code