Support HALF_DAY periodicity for AM/PM date patterns

[seonwooj0810]

Fixes #976

Root cause

The HALF_DAY periodicity type and its RollingCalendar.printPeriodicity() message ("Roll-over at midday and midnight.") have existed since LBCORE-11 (2008), but HALF_DAY was never added to PeriodicityType.VALID_ORDERED_LIST, and RollingCalendar has no HALF_DAY branch in its switch statements. So computePeriodicityType() can never select it: a TimeBasedRollingPolicy whose date pattern carries the AM/PM token but no finer time token (e.g. %d{yyyy-MM-dd-a}) falls through to TOP_OF_DAY and rolls over only at midnight, logging "Roll-over at midnight." instead of rotating at both 00:00 and 12:00.

Change

• PeriodicityType: insert HALF_DAY into VALID_ORDERED_LIST between TOP_OF_HOUR and TOP_OF_DAY. Because detection returns the shortest matching period, this only affects patterns that distinguish 00:00 from 12:00 but nothing finer (i.e. the a token) — patterns that already match at TOP_OF_HOUR or below are unchanged.
• RollingCalendar.innerGetEndOfNextNthPeriod: add the HALF_DAY case — floor to the current half-day boundary (00:00 or 12:00) and advance in 12h steps, mirroring the existing TOP_OF_DAY flooring.
• RollingCalendar.periodBarriersCrossed: add HALF_DAY (previously it hit the default branch and threw IllegalStateException).
• RollingCalendar.isCollisionFree: add HALF_DAY — an isolated a without a date component repeats every 24h, so a 1-day collision check flags it.

Test evidence

Added regression coverage to RollingCalendarTest (all pass; each fails on current master before the fix):

• testPeriodicity: yyyy-MM-dd-a and yyyy-MM-dd a now resolve to HALF_DAY (previously TOP_OF_DAY).
• testVaryingNumberOfHalfDailyPeriods: getEndOfNextNthPeriod advances in exact 12h boundaries across a range of periods.
• testBarrierCrossingComputation: 3 half-day barriers over 36h.
• testCollisionFreenes: yyyy-MM-dd-a is collision-free; isolated a is not.

Verification done: mvn -pl logback-core -am test -Dtest=RollingCalendarTest → 10/10 pass. Ran the full ch.qos.logback.core.rolling.* suite: all rolling tests pass; the only errors are 3 pre-existing InaccessibleObjectExceptions in RollingFileAppenderTest (JPMS reflection into ch.qos.logback.core.appender), which are unrelated to this change and reproduce identically on a clean master checkout when the test JVM is JDK 25.