Polyfill: Replace unreachable retry with assertion in NudgeToCalendarUnit

[jessealama]

The retry path for negative durations appears unreachable: end-of-month constraining reduces the day, which for negative durations pushes the endpoint further from the origin, so the nudge window always contains the destination.

This isn't a rigorous proof of unreachability, but we haven't been able to construct a test case that exercises this path. Replace the retry block with an assertion guarding the invariant, so that if we're wrong, we'll get a clear error rather than silent misbehavior.

Closes #3235

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.00%. Comparing base (d27503d) to head (1221266).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3291      +/-   ##
==========================================
+ Coverage   97.83%   98.00%   +0.16%     
==========================================
  Files          22       22              
  Lines       10725    10711      -14     
  Branches     1856     1855       -1     
==========================================
+ Hits        10493    10497       +4     
+ Misses        215      197      -18     
  Partials       17       17              

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[jessealama]

This change was originally in #3289 (where it was simply deleted, without even an assertion in place).

[ptomato]

I think if we assume this code can't be reached by any JS code, we should change the spec text accordingly to have an assertion there as well, and submit it as an editorial change. (Could possibly also be done after stage 4.)

I have a hunch that the reason the nudge window only needs to be retried when moving forwards, is because the reason for the retry is when the edge of the window lands in a DST gap. The disambiguation: "compatible" behaviour always picks the later time for DST gaps, so that might be why. It's just a hunch, but maybe that helps ruling out why this doesn't happen?

[jessealama]

I think if we assume this code can't be reached by any JS code, we should change the spec text accordingly to have an assertion there as well, and submit it as an editorial change. (Could possibly also be done after stage 4.)

I have a hunch that the reason the nudge window only needs to be retried when moving forwards, is because the reason for the retry is when the edge of the window lands in a DST gap. The disambiguation: "compatible" behaviour always picks the later time for DST gaps, so that might be why. It's just a hunch, but maybe that helps ruling out why this doesn't happen?

I've tweaked the spec text to add an assertion, matching the polyfill code. I've added a note there. I'm happy to open an issue (which this PR would close) where we can try to think out loud about why this behavior is impossible. How does that sound?