revert: retry a hung visual shard on the label and scheduled visual runs

.github/actions/run-visual-shard/action.yml

@@ -15,7 +15,10 @@ inputs:
       How often to run the shard before calling it failed. Firefox
       intermittently hangs on a single file in headless CI — a hang vitest's
       per-test `retry` cannot recover, only a fresh browser process can. A real
-      diff fails every attempt.
+      diff fails every attempt. No caller passes it any more, because the
+      failure it was aimed at is not one a re-run recovers — the attempts share
+      the runner, so all three fail identically (#3106). Kept for a hang that
+      shows up on its own.
     default: "1"
 
 runs:

.github/workflows/test-visual-label.yml

@@ -69,15 +69,10 @@ jobs:
   # browsers, so one runner spent 24-34 minutes on it. Sharding is the only split
   # that keeps each file's conditions identical: every shard still runs its own
   # files one at a time, in its own browser process.
-  #
-  # `attempts` matches the scheduled run, and the timeout matches it for the same
-  # reason: three attempts of a shard have to fit inside the job. This is the
-  # path contributors are told to use, so it is the path that must not report a
-  # contended runner as a diff.
   run-visual-tests:
     needs: prepare
     runs-on: ubuntu-latest
-    timeout-minutes: 45
+    timeout-minutes: 30
     strategy:
       # A failing shard must not cancel the others — the diffs of the shards
       # still running are what tell you whether the change is broad or local.
@@ -98,7 +93,6 @@ jobs:
         with:
           shard: ${{ matrix.shard }}
           shards: ${{ strategy.job-total }}
-          attempts: "3"
 
       # vitest writes the diffs into `.vitest-attachments`, six directory levels
       # deep. Uploaded as they lie, the zip opens on a dot directory Finder does

.github/workflows/test-visual-scheduled.yml

@@ -12,12 +12,15 @@ jobs:
   # styling from any page it is not showing), and it renders every file in both
   # browsers, so one runner spent 24-34 minutes on it. Sharding is the only split
   # that keeps each file's conditions identical: every shard still runs its own
-  # files one at a time, in its own browser process. The retry lives inside the
-  # shard now, so a flaky firefox hang re-runs a sixth of the suite, not all of
-  # it.
+  # files one at a time, in its own browser process.
+  #
+  # No retry: the three attempts this job carried since before the sharding never
+  # recovered the failure they were aimed at. All three fail identically, because
+  # they share the runner — see #3106. The timeout is the label path's, which is
+  # calibrated for this same shape at one attempt.
   visual:
     runs-on: ubuntu-latest
-    timeout-minutes: 45
+    timeout-minutes: 30
     strategy:
       fail-fast: false
       matrix:
@@ -33,7 +36,6 @@ jobs:
         with:
           shard: ${{ matrix.shard }}
           shards: ${{ strategy.job-total }}
-          attempts: "3"
 
   alert:
     needs: visual