Fewer CalendarDay/CalendarDateFromFields calls during PlainYearMonth add/subtract

[arshaw]

While investigating for #2794 (ultimately unsuccessful), I uncovered a way to have less user-code calls from PlainYearMonth add/subtract.

I don't see it in the calendar invariants list, but this axiom probably goes without saying: the day before the first day of one month is the last day of the previous month.

If add/subtract moves backward in time, the algorithm requires us to determine the last day of the current month before proceeding with PlainDate-style add/subtract.

Right now, the last day is being determined, but seems unnecessary to thread it back through CalendarDateFromFields, no?

Adjusted tests:
tc39/test262@main...fullcalendar:test262:temporal-fewer-calls-pym-math

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 96.60%. Comparing base (02a6d14) to head (fe4ad56).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2798      +/-   ##
==========================================
- Coverage   96.60%   96.60%   -0.01%     
==========================================
  Files          23       23              
  Lines       12312    12304       -8     
  Branches     2272     2272              
==========================================
- Hits        11894    11886       -8     
  Misses        356      356              
  Partials       62       62              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[ptomato]

After #2826 we should rewrite this to be an editorial PR. Marking as draft for now.