Add units to output arrays

[Aramist]

Create wrapper types for arrays output by localization models with the following properties:

• Associates a real-world unit (arb., mm, cm, m) and interpretation (Mean, covariance, cholesky) with the array data
• Contains convenience functions for unit conversions

This change aims to resolve the following problems:

• Evaluation and assessment scripts contain many lines of ambiguous shape-checking code to determine what kind of model output was presented (as models can choose to output parametrized densities, mass functions, or point estimates)
• As Numpy arrays and Torch tensors carry no information about units, it is easy to erroneously attempt a unit conversion where none is needed

Merge deadline: Fri 2023-06-16

[amanchoudhri]

In working on this, let's make sure to address #21 by potentially reworking the way model output is saved.