fixes bug #65657, ISO-8601 compliant periods being rejected by parser

[jcomeauictx]

a lot of the parser code is pretty dodgy, and doesn't appear to be using re2c to good advantage, but I focused on fixing the immediate problem with minimal coding changes. I added a test for the bug, verified that it failed before I started coding, and now it passes on both x86_64 and ppc64le virtual hosts.

there is no Makefile rule for creating ext/date/lib/parse_iso_intervals.c from its corresponding .re file, so I ran re2c --no-generation-date -b -o ext/date/lib/parse_iso_intervals.c ext/date/lib/parse_iso_intervals.re. perhaps a rule should be added to Makefile.frag, but I didn't want to dig into how that whole thing works and would rather not step on toes. DRY would suggest having only the .re files in the repository if the .c files are generated from them.

[jcomeauictx]

I see the memory leak in the 2nd travis run. tried to get rid of it by defining char *ZULU_TIME = "Z"; globally and setting t->tz_abbr = ZULU_TIME; but then I get zend_mm_heap corrupted. I'm probably out of my league here. I was never too brilliant about keeping track of pointers and mallocs.

[KalleZ]

Hi

The fix should go to derickr/timelib, where the timelib is maintained, and only contain the test once committed and synced.

cc @derickr

[krakjoe]

As above, please open fresh with test only when upstream has changes ...