Is there a good reason to use `np.float32` throughout `xml_files.py`

[ViridianForge]

This question was raised during the review of #12.

Throughout xml_files.py - all potential floating point data is regarded and cast to np.float32. The issues this presents are two-fold:

1. It adds a dependency of numpy to the class where it may not actually be necessary.
2. It requires np.float32 be cast back into a float for the purposes of serializing the data into json.

The question is - do the reasons for using np.float32 outweigh the issues laid out above?

[ViridianForge]

Two investigatory points to start from.

1 - How much larger do MFFs and JSON that mffpy exports get if we use float in place of np.float32?
2 - What floating point precision do NetStation MFFs utilize?