fix(client): refresh workspace files after disk changes - #4379

💭 Checking convention compliance across the file — error handling uses the proper Schema.TaggedErrorClass with structured attributes and Schema.Defect() for causes, operation discriminators support multiple values with derived messages, namespace imports from effect subpaths are correct, and the service definition follows canonical ordering with proper exports. There's a potential issue emerging with Stream usage that I need to investigate further. I'm noticing that the ws.ts watchFile implementation uses ProjectReadFileError to report watch stream failures, which feels semantically off since it's not really a read-file error. The error contract does declare both ProjectReadFileError and EnvironmentAuthorizationError as possible outcomes, so it's technically consistent with the declared interface. The operation field in the error context includes "watch" to capture what was happening, which provides some semantic clarity even if the error type itself is a bit of a mismatch. Looking at the Effect service conventions in the codebase — the imports are structured properly, there's no misuse of ManagedRuntime or Layer.succeed, and error handling follows the expected patterns with cause chains. The new watch and realpath-watch-directory operations fit naturally into the existing WorkspaceFileSystemOperationError discriminator, and the ws.ts file correctly maps watchFile failures into the error structure.