Windows: every supervised module gets a console window; closing it kills the module (0xC000013A)

[Qiiks]

Upstream creates the module's console here, not at the daemon's own launch.

Two independent reasons this window exists, and only one of them was mine:

1. Upstream gap. spawn_child_in_slot (crates/subc-daemon/src/supervise.rs:5620-5650)
   builds the module Command with piped stdout/stderr and stdin(null), but no
   creation_flags on Windows. A console-subsystem child launched without
   CREATE_NO_WINDOW gets a fresh console, so every supervised module (ck-aft,
   the Synapse module) opens one. Closing that window sends CTRL_C_EVENT, which
   the child reports as 0xC000013A STATUS_CONTROL_C_EXIT — so a user who just
   wants to tidy their desktop kills a healthy module and the supervisor restarts
   it. This is why CREATE_NO_WINDOW exists in setup/uninstall.rs upstream but
   never in supervise.rs.

2. My own regression, in the job-object work. set_suspended_creation_flags
   called command.creation_flags(CREATE_SUSPENDED_FLAG), and
   std::process::Command::creation_flags replaces the whole mask rather than
   OR-ing into it. So a second CREATE_NO_WINDOW call added later would have
   silently dropped CREATE_SUSPENDED — leaving the child both uncontained and
   unresumable. Both flags now go in as one OR'd mask
   (CONTAINMENT_CREATION_FLAGS, crates/subc-jobobject/src/sys.rs).

The fix is one flag at the module spawn site, not at the daemon's own launch
site — the daemon is already launched hidden by the OMP extension
(windowsHide: true), and CREATE_NO_WINDOW is inert for the daemon itself.

The doc comment on SupervisedChild::job already names the remaining
non-Unix gap in this area (drain-then-exit on Windows), which is the separate
follow-up in #111.

[subc-alfonso]

Thanks, the diagnosis is right on both counts. The fix went in with #111, since it's the same flags: modules are now spawned with CREATE_SUSPENDED | CREATE_NO_WINDOW as one mask, set in a single creation_flags call, so neither flag can drop the other. A new Windows test spawns a contained child and checks it has no console window. It reaches a running daemon with the next release.