fix: stop erts when end of input is reached

[clark-lindsay]

This ensures that the runtime will shut down if input is closed, eliminating zombie processes that are left when an explicit "shutdown" message isn't sent to the LS.

I elected to use System.stop/1 instead of System.halt/1 since a careful shutdown of the system seemed preferable.

Testing

I tested the change by replacing the gen_lsp dependency in next-ls with my local version, which includes the change, and comparing the behavior of sending an "End of Transmission" event to a running next-ls instance (started with bin/start --stdio).

Before the change

Running epmd -names | wc -l after starting the server produced 11 (from zombies already on my machine, plus the new server). Without adding the System.stop/1 call and sending an EOT using ^D the server does not stop.

After the change

Running epmd -names | wc -l after starting the server produced 11. Sending an EOT using ^D causes the server to stop, and a subsequent execution of epmd -names | wc -l produces 10.

[mhanberg]

Tested it myself, works with both stdio and tcp mode. I was able to test tcp by doing nc localhost 9000 and then ctrl-c ing out of that, and just for good measure, I had it print out something before it shut down