the gui shell watches files too

It never did — the effect arm was `.watch => {}` with a comment saying a shell
that never delivers the event simply never reloads. That was written as a
deliberate scope cut and it is the whole bug: the user runs pardes-gui.

The tty watcher was fine, and I proved all four suspicions false against a real
binary in a real pty with the write done by a stranger process and NO input
delivered afterwards: the loop does wake (postEvent signals an empty queue),
zig fmt's rename-over does fire MOVED_TO and is caught, three panes across two
directories all reload and closing one leaves its neighbour still following,
and the self-write hash guard does not swallow a real change. The live session
even had its inotify mark on src with the right mask and re-read config.zig
when I touched that directory.

Duplicated rather than shared with tty.zig, the way the two shells already each
own LspJob, forkShell and their pty readers. The headless PARDES_TEST_GRID path
deliberately passes -1: its contract is one frame per scripted input, and a
reload on its own clock would put an unasked-for frame in the stream.

filewatch.snap proved less than it looked. Its real blind spots were the
rename-over shape — the old script only truncated, so it saw CLOSE_WRITE and
never MOVED_TO — multiple directories, and the one that mattered: the suite
only ever runs the TTY binary, so it structurally cannot see a gui-only
regression. It now uses a new `run` directive whose writer is a child of the
RUNNER, so nothing it does reaches an app pty.

1 files changed, 31 insertions, 12 deletions

diff --git a/test/snapshots/filewatch.snap b/test/snapshots/filewatch.snap
index 1ac8d081..6f1f2976 100644
--- a/test/snapshots/filewatch.snap
+++ b/test/snapshots/filewatch.snap
@@ -1,7 +1,23 @@
 # external file updates: a write from outside pardes lands in the open pane,
 # and Undo brings the unsaved edit back. That is the whole contract — the
 # update is committed like any other edit, so nothing has to be merged.
-lines w.txt 12
+#
+# The writer must be a STRANGER to the editor. This script used to have a
+# pardes PANE run the write, which proved much less than it looked: that
+# shell's own pty output wakes the loop and paints a frame, so the reload
+# could be riding on the keystroke's frame instead of on the watch. `run` is a
+# child of the RUNNER — nothing it does reaches an app pty — and every
+# assertion below it is a `wait`, which only READS the pty. So between the
+# write and the reload the app receives no keystroke, no resize, no mouse: if
+# the watch does not wake the loop by itself, these waits time out.
+#
+# Two files in two directories, written the two ways files actually change:
+# w.txt in place (CLOSE_WRITE) and sub/s.txt by rename-over, which is what
+# `zig fmt`, `git checkout` and every editor's atomic save do (MOVED_TO, and
+# a new inode — the reason the mark is on the directory).
+dirmk sub
+file w.txt line 1\nline 2\n/tmp/pardes-snap/filewatch/cwd/sub/s.txt\nline 4\nline 5\nline 6\nline 7\nline 8\nline 9\nline 10\nline 11\nline 12\n
+lines sub/s.txt 8
 start 30 150 -n 3
 wait 8000 w.txt
 stable 700 20000
@@ -15,9 +31,15 @@ key esc
 settle 100
 press right 6 8
 release right 6 8
-wait 10000 line 3
+wait 10000 line 4
 stable 700 15000
-# an UNSAVED edit on line 1
+# ...and open sub/s.txt from the path sitting on line 3 of that pane, so the
+# second watched pane is in a DIFFERENT directory from the first
+press right 20 5
+release right 20 5
+wait 10000 line 8
+stable 700 15000
+# an UNSAVED edit on line 1 of w.txt
 press left 10 3
 release left 10 3
 stable 400 5000
@@ -27,18 +49,15 @@ key esc
 settle 100
 stable 400 5000
 snap edited
-# overwrite the file from a shell in another column. The marker is split so
-# the ECHOED command line never matches the wait — only the reloaded pane does
-press left 60 5
-release left 60 5
-stable 400 5000
-key c-b
-stable 600 8000
-text echo EXTERN''AL > w.txt
-key enter
+# ---- from here to `snap reloaded_sub`, pardes is sent NOTHING ----
+run printf 'EXTERNAL\n' > w.txt
 wait 10000 EXTERNAL
 stable 700 10000
 snap reloaded
+run printf 'RENAMED\n' > sub/t && mv sub/t sub/s.txt
+wait 10000 RENAMED
+stable 700 10000
+snap reloaded_sub
 # back in the file pane: one undo and the unsaved edit is there again
 press left 10 3
 release left 10 3