Add support for changing the year when starting a branch run

A user ran 9982 years of a 3 degree ocean-ice run (the CORE forcing dataset covers 62 years, so this represents 161 cycles). They then tried to branch off the run for a couple of experiments, and ran into an `illegal linear times` issue when 9999-12-31 tried to roll over to 10000-01-01. (I believe he's using CESM 2.1, which is MCT-based rather than NUOPC-based, so the error message out of a recent beta tag might be different.)

Anyway, setting up a `branch` run instead of a `hybrid` is preferable because a hybrid run doesn't start the ocean component until the second coupling interval... so a branch would be a bit-for-bit continuation while a hybrid would change answers. However, he currently has to use a hybrid because the branch forces the user to keep the calendar from the run being continued.

While I could see an issue with creating a branch run B from the January 1st restart files from run A and trying to claim B starts in the middle of the year, a way to either reset the year to 0001 or set it to an arbitrary YYYY value while keeping the month and day constant would be helpful.

At a recent CSEG meeting, we discussed another option for supporting this use case: using 5 or 6 digits for the year instead of the current 4 digits. That seems more complicated than letting users branch from 9999 and reset the calendar back to 0001, and also has the potential for creating other issues (such as https://github.com/ESMCI/cime/issues/2903).