This supercedes other issues that I will link and close here: #49, #50, #51

The core idea is that a vocalpy.segment function can return one of two things:

1. Segments, a set of line segments. For multiple reasons (detailed in linked issues) we choose to represent this segmentation $S$ of a digital signal $y[n]$ with $N$ samples as a series of $m$ consecutive, non-overlapping segments $s_i$ where each $s_i$ is a tuple with a start time (an integer) index and a duration (length in number of samples) $(t_i, d_i)$
2. Boxes, a set of boxes ... where I don't have a particularly strong opinion about the representation just yet.

Importantly, both Segments and Boxes can have a Sound and Spectrogram instance associated with each element of the set. This gives us a way to extract features directly from a segmentation, without needed to somehow use the segmentation to index back into a Sound or Spectrogram