feat(kubernetes): cluster state and kubeconfig watchers

[manusa]

Follows up on #502 to better organize the code and improve its maintenance and sustainability.

Changes:

• Extract watchers into new pkg/kubernetes/watcher package with a common Watcher interface
• Move watcher ownership from Manager to Providers (kubeConfigClusterProvider, singleClusterProvider)
• Simplify Manager by removing watch-related fields and methods (continues the work on refactor(kubernetes): enhance AccessControlClientset and streamline Manager and Kubernetes structs #503)
• MCP server refactor:
  • Creates the provider instance only once (maybe this will need changing depending on the scope of what needs to be done for feat(kubernetes): add cluster state monitoring and drop-in config support and SIGHUP reload #493)
  • Exposed a reloadToolsets method that is used as the watcher callback.
• Extract IsOpenshift() into pkg/openshift package for reuse
• Add Manager.Invalidate() method for cache invalidation
• Add environment variable overrides (CLUSTER_STATE_POLL_INTERVAL_MS, CLUSTER_STATE_DEBOUNCE_WINDOW_MS) for testing

[manusa]

For f54d30e

IMO we should actually move the SIGHUP logic to the pkg/kubernetes-mcp-server/cmd package where it makes more sense.

[nader-ziada]

i like the moving of watcher logic in a separate package, much cleaner. Also the move of open shift check. 👍

/lgtm

[nader-ziada]

thanks for the refactor @manusa, i will rebase the drop-in config and make the suggested changes

/lgtm
/approve