evaluation based on phase_of_evolution

Hi all, more `phase_of_evolution` content:
I ran into serious issues trying to use `s% phase_of_evolution >= 2` as a condition for activating a particular modification within `other_MLT_results` in `run_star_extras` (this is related to my contextless post about `retry: error in do1_dlnT_dm_eqn 3` in bugs-and-problems on Slack). When running an 0.2Ms model at solar composition,

this (case 1) runs fine:
```
if (s% phase_of_evolution >= 3) then
gradr_spot = gradr/factor
else
gradr_spot = gradr
end if
```

as does this (case 2):
```
if (s% star_age >= 1d0) then ...
```

but this (case 3) does not:
```
if (s% phase_of_evolution >= 2) then
gradr_spot = gradr/factor
else
gradr_spot = gradr
end if
```
even though `s% star_age = 1 ! yr` is definitely after the onset of `phase_of_evolution = 2` (the transition time for this parameter combination between phase 1 and phase 2 is about 1d-5 yr)
[error.log](https://github.com/MESAHub/mesa/files/7471479/error.log)

"factor" is a number, gradr_spot is the same type as MLT's standard gradr ( type(auto_diff_real_star_order1) )

The result of running case (3) is attached, but can be summarized by
`retry: error in do1_dlnT_dm_eqn` a bunch of times until `failed in do_relax_num_steps`.

I confirmed that `phase_of_evolution = 1` is met before immediately crashing at `phase_of_evolution = 2`. I suspect that things work exactly one timestep after the `phase_of_evolution = 2` switch comes on, but not _at_ the switch. Tagging @adamjermyn at his request

Reproduce the three cases in run_star_extras, focusing on the phase_of_evolution condition inside other_MLT_results. Start with the attached error.log and the retry: error in do1_dlnT_dm_eqn and failed in do_relax_num_steps messages, then compare behavior at the phase 1-to-2 transition. Done means case 3 no longer crashes at that switch.

Written by the indexing model from the issue text.