perf(jsonrpc): buffer output writes in StreamMessageConsumer

[sebthom]

Wrap provided OutputStream in BufferedOutputStream to amortize small writes; improves throughput by reducing syscalls. Centralize wrapping via setOutput to avoid double wrapping.

[jonahgraham]

This PR contains 3 different changes:

1. synchronizing changes of output on outputlock (I assume so that output can't change in the middle of a message)
2. fixing test suite by using atomic string
3. wrapping output in buffered output stream.

1 and 2 are in principle fine - 3 has an API problem, the getOutput method now unexpectedly no longer returns what I passed in to setOutput or the constructor.

I also don't really like the need to lock to change the output - I think the output should be unchangeable. LSP4E certainly doesn't seem to change it. Not sure of others.

I would be inclined for LSP4J 1.0.0 to remove the public setOutput/getOutput and then simplify this code.

[sebthom]

@jonahgraham maybe you could create another 0.x release where the getOutput/setOutput are deprecated and marked for removal.

[jonahgraham]

@jonahgraham maybe you could create another 0.x release where the getOutput/setOutput are deprecated and marked for removal.

Major releases are API breaking. I don't think we need to deprecate it ahead of time. See https://github.com/eclipse-lsp4j/lsp4j/blob/main/README.md#published-api-changes (just merged from #921).