Improve code quality: graceful shutdown and concurrency safety

[hajnalmt]

Context

The codebase has several code quality problems that should be addressed to bring it up to modern Go standards and make it production-ready for Kubernetes environments.

Problems

No Graceful Shutdown

The main loop runs forever with no signal handling. There is no way to catch SIGTERM/SIGINT for clean shutdown, cancel in-flight operations, or log shutdown events. This is problematic in Kubernetes where pods receive SIGTERM before being killed.

Global Mutable State Without Synchronization

• numaMap (package-level map) is read and written without mutex protection
• config (package-level var) is similarly unprotected
• While currently single-goroutine, this is fragile and prevents future concurrency (e.g., adding an HTTP health endpoint)

Hardcoded Strings

• Environment variable names like MY_NODE_NAME are used as magic strings in multiple files

Tasks

• Implement graceful shutdown with signal.NotifyContext and context propagation
• Add sync.RWMutex protection for package-level mutable state
• Extract hardcoded strings into package-level constants
• Improve error messages for missing configuration

Acceptance Criteria

• Binary handles SIGTERM gracefully (logs shutdown, flushes, exits cleanly)
• No data races detected by go test -race ./...
• golangci-lint run ./... passes with zero issues

Priority

Medium — important for production Kubernetes deployments where pods receive SIGTERM during rolling updates, node drains, and scaling events.