Thank you for your interest in improving Matplotlib!

However, this portion of that issue has already received a PR (#27494).
In general, barring significant technical advantages, we tend to prefer to give the first author to open a PR the opportunity to finish their contribution.

And in looking into it we discovered that the tri* plots are not actually properly handling unit-ful inputs, so we raised #27499. Your plot is hiding these failures because it a) calls plot so the axis units get set, when they are not properly set by tricontour alone (but should be) and b) happen to "work" because they use datetime64[D] dtype which coincides with matplotlib's floating point representation of dates for axes, but it does not actually pass through the units code so if you used a different scale (e.g. datetime64[s]) then you would get the wrong thing plotted. (or indeed changed the epoch date for mpl from 1970-01-01)

Alright, thanks for the quick feedback!