common: guard the reset-path re-fetch in sync_clone so an offline blip still exits EX_TEMPFAIL under set -e

The reset-failure recovery path called journal_fetch as a bare statement
(journal_fetch "$dir"; rc=$?). When the hard reset fails AND the re-fetch
also fails (the classic connectivity outage), the bare call is itself a
set -e exit: every set -e caller (claim-job, complete-job, all producers)
died with the raw fetch rc before _fetch_stderr_is_offline could classify
the blip and take the clean exit $GARDEN_OFFLINE_RC (75) tick-skip —
re-introducing the exact per-blip FATAL the first fetch path was hardened
against. Capture the rc through an if, the same guarded idiom as the
first fetch.

Repro (bare set -e caller; fetch#1 ok, reset fails, re-fetch offline):
BEFORE exit rc=128 raw, no classification; AFTER exit 75 with
"offline on reset; skipping tick". fetch-timeout-test.sh 12/12 and
stale-lock-test.sh 7/7 pass.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Author: endolinbot

scripts/jobs/common.sh

@@ -2001,7 +2001,11 @@ sync_clone() {
   # a connectivity outage, so exit EX_TEMPFAIL exactly like the fetch path. A
   # reset that fails for any other reason still surfaces (the retry below dies).
   if ! git -C "$dir" reset -q --hard "origin/$JOURNAL_BRANCH"; then
-    journal_fetch "$dir"; rc=$?
+    # Same guarded idiom as the first fetch above: a bare `journal_fetch ...; rc=$?`
+    # is a `set -e` exit at the call itself when the re-fetch ALSO fails (the classic
+    # connectivity outage), killing the process with the raw rc before the offline
+    # classification below can run.
+    if journal_fetch "$dir"; then rc=0; else rc=$?; fi
     if [ "$rc" -ne 0 ] && _fetch_stderr_is_offline "$GARDEN_FETCH_STDERR"; then
       log "offline on reset; skipping tick (rc=$GARDEN_OFFLINE_RC)"
       exit "$GARDEN_OFFLINE_RC"