Measured on .173: the guard never fired in the one case it exists for. After a
desktop-mode session disconnects its monitor LINGERS, so `no_live_displays()`
— which counted keep-alive slots as held — refused the restart on exactly the
reconnect that needed it, and the capture session went back to
`cursor_excluded=true` and forced compositing.
Only a streaming session (SlotState::Active) can be damaged by the restart. A
lingering/pinned monitor has no session attached and a reconnect preempts and
recreates it anyway ("a reused IddCx swap-chain is dead"), so the restart
destroys nothing that was going to survive.
enricobuehler
b20184462d