Fix rotation and ST time smoothing state on restart

[Debraheem]

Photo restarts could use an uninitialized w_div_w_crit_roche while reconstructing i_rot. They could also discard mstar_old and the saved ST coefficients before the first post-restart remeshing event.

The solution is to Initialize the rotation geometry before reconstructing i_rot, save mstar_old in photos, and preserve the ST time smoothing inputs through the first new_generation. Bump the star photo version to 19. Part of the solution uses the photo restart fixes to ST taken from https://github.com/MESAHub/mesa/pull/553/changes (originally found by @rjfarmer)

The three affected star test cases pass their restart and checksum checks with FPE checks enabled:

• accreted_material_j
• check_redo
• high_rot_darkening

The ST restart state was also verified with double_bh through a model-700 restart, but it also revealed a second fpe issue in the binary Roche potential routines. That problem is specific to nans and -infinity values inside the Roche potential tables and deserves its own separate pr.

[Debraheem]

the double_bh fpe failure on my machine is a known issue related to fpe checks complaining about the nans in the roche tables. Otherwise, the tests are passing and the fpe checks passed on BOSE.