Do not format USEC_INFINITY as NULL

systemctl would print 'CPUQuotaPerSecUSec=(null)' for no limit. This
does not look right.

Since USEC_INFINITY is one of the valid values, format_timespan()
could return NULL, and we should wrap every use of it in strna() or
similar. But most callers didn't do that, and it seems more robust to
return a string ("infinity") that makes sense most of the time, even
if in some places the result will not be grammatically correct.

src/test/test-time.c

@@ -43,12 +43,18 @@ static void test_parse_sec(void) {
         assert_se(u == 2500 * USEC_PER_MSEC);
         assert_se(parse_sec(".7", &u) >= 0);
         assert_se(u == 700 * USEC_PER_MSEC);
+        assert_se(parse_sec("infinity", &u) >= 0);
+        assert_se(u == USEC_INFINITY);
+        assert_se(parse_sec(" infinity ", &u) >= 0);
+        assert_se(u == USEC_INFINITY);
 
         assert_se(parse_sec(" xyz ", &u) < 0);
         assert_se(parse_sec("", &u) < 0);
         assert_se(parse_sec(" . ", &u) < 0);
         assert_se(parse_sec(" 5. ", &u) < 0);
         assert_se(parse_sec(".s ", &u) < 0);
+        assert_se(parse_sec(" infinity .7", &u) < 0);
+        assert_se(parse_sec(".3 infinity", &u) < 0);
 }
 
 static void test_parse_nsec(void) {
@@ -125,6 +131,7 @@ static void test_format_timespan(usec_t accuracy) {
         test_format_timespan_one(986087, accuracy);
         test_format_timespan_one(500 * USEC_PER_MSEC, accuracy);
         test_format_timespan_one(9*USEC_PER_YEAR/5 - 23, accuracy);
+        test_format_timespan_one(USEC_INFINITY, accuracy);
 }
 
 static void test_timezone_is_valid(void) {