Fix potential deadlock in case of lsp shutdown

[fubhy]

The LSP server could deadlock during shutdown when background tasks or handlers attempted to send messages (logs, diagnostics, responses) to the outgoingQueue after the writeLoop had exited. With no consumer draining the queue, these blocking channel sends would previously hang forever.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[jakebailey]

Thanks for looking into this. We really shouldn't have code that actually uses Background() if it's long running or interacts like this; there should be a path to plumb a proper context, I would think.

[fubhy]

There should be a path to plumb a proper context, I would think.

Along these lines ... ?

[fubhy]

The latest CI error is again caused by this: #2485

[jakebailey]

I don't know how you're hitting these races in CI so often when I don't even remember seeing them anywhere until now 😅

[fubhy]

I don't know how you're hitting these races in CI so often when I don't even remember seeing them anywhere until now 😅

Hah I guess that's just because I'm working on this in so many incremental steps because I'm so inexperienced with this codebase (and Go, frankly) that I trigger more CI runs than you maybe? ;-)

[jakebailey]

I feel like this is good, now.