core: serialize and deserialize unit start ratelimits

The logic is taken from dump ratelimit: if the config changes, we discard the
counters. This allows the user apply new limits and "start from scratch" in
that case.

This actually makes StartLimitIntervalSec=infinity (or with a large interval)
work as expected, because the counter is maintained even if daemon-reload
operations are interleaved.

test/units/testsuite-04.SYSTEMD_JOURNAL_COMPRESS.sh

@@ -7,6 +7,9 @@ set -o pipefail
 mkdir /run/systemd/system/systemd-journald.service.d
 MACHINE_ID="$(</etc/machine-id)"
 
+# Reset the start-limit counters, as we're going to restart journald a couple of times
+systemctl reset-failed systemd-journald.service
+
 for c in NONE XZ LZ4 ZSTD; do
     cat >/run/systemd/system/systemd-journald.service.d/compress.conf <<EOF
 [Service]
@@ -35,4 +38,5 @@ done
 rm /run/systemd/system/systemd-journald.service.d/compress.conf
 systemctl daemon-reload
 systemctl restart systemd-journald.service
+systemctl reset-failed systemd-journald.service
 journalctl --rotate

test/units/testsuite-67.sh

@@ -88,6 +88,8 @@ do
     # Make sure we are in a consistent state, e.g. not already active before we start
     systemctl stop systemd-integritysetup@"${DM_NAME}".service || exit 1
     systemctl start systemd-integritysetup@"${DM_NAME}".service || exit 1
+    # Reset the start-limit counters, as we're going to restart the service a couple of times
+    systemctl reset-failed systemd-integritysetup@"${DM_NAME}".service
 
     # Check the signature on the FS to ensure we can retrieve it and that is matches
     if [ -e "${FULL_DM_DEV_NAME}" ]; then