Move profile collection logic under shutdown

[radeksimko]

Background

Whenever users report memory usage related issues, we always advise them to collect a memory profile.

Currently we collect the profiles upon cancellation (typically triggered by exit RPC notification and/or SIGINT).

terraform-ls/internal/cmd/serve_command.go

Lines 71 to 82 in 81b49a1

	if c.cpuProfile != "" {
	stop, err := writeCpuProfileInto(c.cpuProfile)
	defer stop()
	if err != nil {
	c.Ui.Error(err.Error())
	return 1
	}
	}
	if c.memProfile != "" {
	defer writeMemoryProfileInto(c.memProfile)
	}

While many users were able to obtain profiles that way, we discovered this strategy has two problems:

• writing profile files is not guaranteed to be quick, which may get the client to SIGKILL the process while it's writing to disk - see Avoid killing server before it had chance to shut down gracefully sublimelsp/LSP#1660 for more context
• Windows seems to have different ways of shutting down processes which seems to prevent us from catching the signals we normally do on Unix. See terraform-ls.exe uses ~500 MB of RAM for each vscode window #986 (comment)

terraform-ls/internal/cmd/serve_command.go

Lines 98 to 100 in 81b49a1

	ctx, cancelFunc := lsctx.WithSignalCancel(context.Background(), logger,
	os.Interrupt, syscall.SIGTERM)
	defer cancelFunc()

Proposal

Move profile file to the shutdown handler

terraform-ls/internal/langserver/handlers/shutdown.go

Lines 7 to 9 in 81b49a1

	func Shutdown(ctx context.Context, _ interface{}) error {
	return nil
	}