Improve cache coherence

[jvansanten]

The northern tracks data sample comes in a giant structured array with some unused columns. This is inefficient in several ways:

• Reading the array from disk takes several seconds, as even the columns that are not used have to be read
• Reading an individual column makes poor use of caches, as adjacent values in the column are separated by strides that are large compared to a cache line

Improve both of these by reading data from an npz file if present (unused columns are never read), and storing data column wise in astropy tables.

[jvansanten]

Close in favor of #444