fix: prevent server process leak (#154)

[pranaygp]

Summary

• Closes electron-nodejs (server.js) Spamming #154 (150+ electron-nodejs server.js processes, ~40-67MB each).
• didOpenTextDocument used !clients.has(folder.uri.toString()) to avoid duplicate spawns, but inserted into clients only inside the .then() of an async Workspace.findFiles(...). Every document opened between the findFiles call and its resolution passed the guard and started a new LanguageClient for the same workspace folder. At activation, Workspace.textDocuments.forEach(didOpenTextDocument) invokes this synchronously per open document, producing one server.js process per supported document — matching the reporter's screenshot.
• Track in-flight client creation in pendingClientFolders so concurrent calls coalesce, and clear that set on workspace folder removal so a folder closed mid-findFiles doesn't leave a stranded server.

Test plan

• yarn build passes
• yarn test passes (24/24)
• Manual verification: run VS Code with a workspace containing many CSS/SCSS/LESS files (or many files in peekFromLanguages) — confirm exactly one server.js process is spawned per workspace folder, not one per document. Toggling workspace folders should add/remove processes 1:1.

Manual verification of the leak requires running VS Code for an extended period with workspaces containing many supported documents; there is no autonomous repro in the test suite.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Fixes a client/server process leak in the VS Code extension by preventing multiple LanguageClient instances from being spawned concurrently for the same workspace folder during activation and rapid document opening.

Changes:

• Adds pendingClientFolders to coalesce concurrent didOpenTextDocument calls while Workspace.findFiles(...) is in flight.
• Cancels in-flight client creation when a workspace folder is removed to reduce the chance of spawning servers for closed folders.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[pranaygp]

Fixed in d0791c6. Reordered so clients.set(folderKey, client) and pendingClientFolders.delete(folderKey) now happen before client.start() — so an onDidChangeWorkspaceFolders removal racing with this code can find the client in clients and stop it instead of leaking the server process. Wrapped start() in a try/catch: on failure we clients.delete(folderKey) (pending was already cleared) so the folder can be retried later, and we log via reporter.sendTelemetryErrorEvent.

[pranaygp]

Fixed in d0791c6. Replaced the rethrow in the findFiles rejection handler with reporter.sendTelemetryErrorEvent(err, { context: "client", method: "startClientForFolder" }) and an implicit return. Pending state is still cleared first, so the folder can be retried later. This avoids the unhandled promise rejection — the surrounding try/catch couldn't catch it since the .then isn't awaited.