process: kill on drop (meta-pytorch#802)

Summary:
Pull Request resolved: meta-pytorch#802

the test added in D77348271 not withstanding, child process cleanup is evidently unreliable after D77392241 as evidenced by running `buck2 test fbcode//monarch/hyperactor_mesh:hyperactor_mesh` and observing that on completion, 100+ orphaned instances of `hyperactor_mesh_test_bootstrap` are left running.

this diff implements the only way i've found that fixes this problem (even reintroducing parent death detection via signal handling does not leading me to conclude it never did and that the line this diff puts back is in fact, the reason it seemed to).

ghstack-source-id: 301707897
exported-using-ghexport

Reviewed By: mariusae

Differential Revision: D79898223

fbshipit-source-id: 74001e5a1a763bd264acb7518f7970d2ccdc1478

Author: shayne-fletcher

hyperactor_mesh/src/alloc/process.rs

@@ -371,6 +371,7 @@ impl ProcessAlloc {
         cmd.env(bootstrap::BOOTSTRAP_LOG_CHANNEL, log_channel.to_string());
         cmd.stdout(Stdio::piped());
         cmd.stderr(Stdio::piped());
+        cmd.kill_on_drop(true);
 
         let proc_id = ProcId::Ranked(WorldId(self.name.to_string()), index);
         tracing::debug!("Spawning process {:?}", cmd);