Adding torch.tensor(np.nan) bug fix

[jennmald]

Adds the fix for torch.tensor(np.nan) for #76 in agent.py
Still needs test case for this fix using Ophyd sim and the updated doc string to explain the enforce_all_objectives_valid parameter.

Closes #76

[jennmald]

@mrakitin @yxrmz I made the fix for the open issue but am having trouble coming up with a test case still using Ophyd sim. Let me know if either of you have ideas. thanks!

[mrakitin]

@thomaswmorris, could you please help Jennefer with the description of the parameter?

[thomaswmorris]

@jennmald you would turn on enforce_all_objectives_valid if you want to throw away all the beam diagnostics when one of them is bad, e.g. if the flux is invalid because it's too low and so you want to automatically throw away the size estimate. But this isn't always true, e.g. at TES where you can measure the flux separately from the size and so we would still want to fit to the flux even if the size diagnostic is messed up.

[jennmald]

Thanks @thomaswmorris !