Improve test coverage for AdjustRoundedDurationDays

[arshaw]

The AdjustRoundedDurationDays function has special behavior for balancing-up time units during a ZDT days:

proposal-temporal/polyfill/lib/ecmascript.mjs

Lines 5262 to 5269 in 2c9979f

	// There's one more round of rounding possible: if relativeTo is a
	// ZonedDateTime, the time units could have rounded up into enough hours
	// to exceed the day length. If this happens, grow the date part by a
	// single day and re-run exact time rounding on the smaller remainder. DO
	// NOT RECURSE, because once the extra hours are sucked up into the date
	// duration, there's no way for another full day to come from the next
	// round of rounding. And if it were possible (e.g. contrived calendar
	// with 30-minute-long "days") then it'd risk an infinite loop.

FullCalendar's temporal-polyfill has an incorrect implementation of this not caught by tests. It'd be great to improve test coverage.

dur = Temporal.Duration.from({ hours: 13 })
zdt = Temporal.ZonedDateTime.from('2024-03-10T00:00:00[America/New_York]') // a 23-hour day
roundedDur = dur.round({
  relativeTo: zdt,
  largestUnit: 'years',
  smallestUnit: 'hours',
  roundingIncrement: 12,
  roundingMode: 'ceil'
})
// EXPECTED: { days: 1, hours: 12 }

Things work fine in the reference implementation.

[catamorphism]

I'm working on this, but I think it should wait until after #3036 lands; otherwise things get too confusing. (Never mind, it's fine.)

[ptomato]

I wonder if this case should've been included in #2916 and therefore disallowed. But on the other hand, with rounding vs 23-hour days it doesn't seem to be as confusing as the cases in #2916.

[ptomato]

Coverage added in test262.