The fire-and-forget run captured its session's pi/ctx; after /new, fork,
switch, or reload pi invalidates those, so the orphaned run's notify/
sendMessage threw 'This extension ctx is stale after session replacement
or reload'. Now session_shutdown aborts the in-flight run before
invalidation (pi.exec supports AbortSignal), and stale errors are
swallowed as a race guard.
Also show a 'running tests...' footer status widget while the run is
in flight, cleared on completion.
Firehose Bot
fc2bafdfa1