feat(language-server): Support client-side file watching via onDidChangeWatchedFiles

This implements `onDidChangedWatchedFiles` in the language server, which
allows the client to communicate changes to files rather than having the
server create system file/directory watchers.

This option is enabled in the extension via the
`angular.server.useClientSideFileWatcher` setting.
When enabled, the extension registers a FileSystemWatcher for .ts, .html, and package.json files and forwards events to the server. The server completely disables its internal native file watchers (via a new 'ServerHost' implementation that stubs watchFile/watchDirectory).

This is significantly more performant and reliable than native watching for several reasons:
- Deduplication: VS Code already watches the workspace. Piggybacking on these events prevents the server from duplicating thousands of file watchers.
- OS Limits: Since the server opens zero watcher handles, it is impossible to hit OS limits (ENOSPC), no matter how large the repo is.
- Optimization: VS Code's watcher uses highly optimized native implementations (like Parcel Watcher in Rust/C++) which handle recursive directory watching far better than Node.js's 'fs.watch'.
- Debouncing: The client aggregates extremely frequent file events (e.g., during 'git checkout'), reducing the flood of processing requests to the server.

This option was tested in one very large internal project and observed
~10-50x improvement of initialization times.

fixes #66543

vscode-ng-language-service/package.json

@@ -153,6 +153,11 @@
           "type": "string",
           "default": "",
           "markdownDescription": "A comma-separated list of error codes in templates whose diagnostics should be ignored."
+        },
+        "angular.server.useClientSideFileWatcher": {
+          "type": "boolean",
+          "default": true,
+          "markdownDescription": "When enabled, the Angular Language Service will delegate file watching to VS Code instead of creating its own internal file watchers. This can significantly improve performance (greater than 10x faster initialization) and reduce resource usage in large repositories."
         }
       }
     },