Make loading IRISSpectrograph more memory efficient

[nabobalis]

I'm finding that IRISSpectrograph, or read_iris_spectrograph_level2_fits() takes a very long time to load and has a very high memory usage. Loading a moderate size IRIS raster in my laptop, which only has 8 GB of RAM, leads to lots of swapping and the machine nearly crawls to a halt.

Testing in different machines, I find the RAM that read_iris_spectrograph_level2_fits() uses is between 3-4 times the total size of the IRIS level 2 FITS file. A 606 MB file took about 1.5 GB of RAM, while a 2.6 GB file took 10GB. This is very inefficient, not to mention slow (even if the machine has enough RAM).

Looking in the code, I suspect these lines are the most problematic:
https://github.com/sunpy/irispy/blob/bbd194205c1dd815230925ad676a286f572bb28c/irispy/spectrograph.py#L562-L564

There should be a way to load it with memmap, and ideally it should be the default.

In addition, the uncertainty will perform many operations on all data points from the file:

https://github.com/sunpy/irispy/blob/bbd194205c1dd815230925ad676a286f572bb28c/irispy/spectrograph.py#L596-L598

This is a huge burden for something of marginal value.

The loading of IRISSpectrograph should be as lightweight as possible. People may use it just to look at what's inside.

Original issue: sunpy/sunraster#82