Fix 12am and 12pm being swapped in parseATTime

[dylanpulver]

webapp/graphite/render/attime.py:84,90,96 maps an am hour straight through and a pm hour through (hour + 12) % 24. That is correct for 1 through 11 and inverted at 12:

12am    -> 2020-03-15T12:00:00+00:00     (expected 00:00)
12:30am -> 2020-03-15T12:30:00+00:00     (expected 00:30)
12pm    -> 2020-03-15T00:00:00+00:00     (expected 12:00)
12:30pm -> 2020-03-15T00:30:00+00:00     (expected 12:30)

so &from=12am&until=12pm graphs noon → midnight — a different 12-hour window, silently, with no error.

docs/render_api.rst:188 calls the format at(1)-compatible, and POSIX at(1) defines midnight as "the time 12:00 am (00:00)" and noon as "the time 12:00 pm". The same module already resolves noon to 12:00 and midnight to 00:00, so today one request can get two different times for the same instant depending on which spelling is used.

Measured with parseATTime(ref, pytz.utc, now=2020-03-15 09:30) over midnight, 12am, 12:30am, 1am, 11am, noon, 12pm, 12:30pm, 1pm, 11pm, teatime: 4/11 wrong before, 0/11 after; the seven controls pass on both sides.

webapp/tests/test_attime.py never used hour 12 — 8:50am, 8:50pm, 8am, 10pm — and 12 is the only hour where (h + 12) % 24 is wrong, so the whole suite passed. Four cases added, in both parseTimeReferenceTest and parseTimeReferenceTestNow.

Mutants: reverting the source leaves exactly the 6 new tests failing and nothing else; fixing only the bare Xam/Xpm branches leaves 12:30am/12:30pm failing; fixing only the pm side leaves the three 12am cases failing.

manage.py test, Python 3.13 (in the CI matrix): 859 tests / 18 errors on clean master, 865 tests / 18 errors with this change — the 18 are pre-existing module-import errors from optional dependencies I don't have installed (whisper, ceres, rrdtool, redis), identical on both sides.

Not tested: any hour above 12 with an am/pm suffix. (h + 12) % 24 and h % 12 + 12 disagree there (13pm was 01:00, is now 13:00), but POSIX at restricts wallclock_hour to 1–12 so both are garbage-in. I also have not run the mysql/postgresql tox targets or anything on Windows.

Written with Claude Code. On provenance: this did not come out of a production incident. I was checking attime.py's hour handling against the at(1) semantics the docs claim, hour by hour, and 12 is where it breaks.