Check for zero-valued alpha parameter

[eirrgang]

It doesn't make sense for alpha to be zero unless a developer is testing a restraint that should not apply forces. However, under some circumstances, run_brer could accidentally set alpha=0. We should check for this and either issue a warning or raise an error that can be optionally disabled.

[eirrgang]

There are two places we can deal with this, and I suggest we use both.

1. The Builder for the BRER restraint plugin instance does not accept an initialization argument for alpha (or alpha_prev or alpha_max).
   At

   brer_plugin/src/cpp/brerpotential.h

   Line 30 in dc91b0b

   double alpha{0};

   they are all (always) initialized to 0. We could choose a small but sensible alternative value. (1.0?)

2. In run_brer, before https://github.com/kassonlab/run_brer/blob/07872e26abcb3e9357efa7acd70ff395f03a8c11/run_brer/plugin_configs.py#L118 or https://github.com/kassonlab/run_brer/blob/07872e26abcb3e9357efa7acd70ff395f03a8c11/run_brer/plugin_configs.py#L147 we can check for a zero-valued alpha and raise a warning or error. (We could make the error conditionally fatal, checking for sys.flags.debug.)

update Option 3: Don't report alpha until the potential is reporting that it has completed training. I think this is logically a separate (but related) issue / feature.

[eirrgang]

run_brer now checks for zero-valued alpha and raises an exception.

It is reasonable to initialize the struct with alpha{0}.

If anything, we could add a check to the plugin interface that requires an explicit alpha and checks whether it is zero.

But none of us seem to think this is necessary, given the updated checks in run_brer and the ability to retrieve the alpha value from the plugin instance.