WIP: bugfixes for GST with penalties in objective functions

[rileyjmurray]

@coreyostrove reported bugs in fitting GST models where the objective function contains penalty terms. Quoting from an email:

The functionality in question is the ability to add in modifications to the objective functions in the form of a penalty function. The one relevant to this discussion which we were trying to test out for gauge regularization purposes is an existing one set using the 'errorgen_penalty_factor' option. This dies somewhere in customsolve, as I recall. For the sake of testing whether the issue was isolated to this particular penalty function we also tried the built-in 'cptp_penalty_factor' and 'spam_penalty_factor' which also both failed (though in different ways for each option).

Running a notebook that Corey sent me, I got the following error. The stacktrace isn't shown, but the error indeed happens when adding penalty terms to the Jacobian, inside simplerlm.py::simplish_leastsq and/or customlm.py::custom_lstsq.

I was able to correct the error above by adding input handling to vec_to_stdmx, casting non-numpy objects to ndarrays. (Currently the only change in this PR.) Of course, we really need to find the functions that are calling vec_to_stdmx with this invalid input. The full extent of such functions is unknown (to me) at time of opening this PR. It at least includes objectivefns.py::_spam_penalty_jac_fill.

Opening this PR at @coreyostrove's request. CC: @pcwysoc.

[rileyjmurray]

@pcwysoc you sent an email asking where the penalty fix(es) are. Here's the WIP PR!