Add ability to plot waterfalls of delay spectra

[philbull]

Plot delay spectra ordered by time or baseline-pair. Solves #155.

[coveralls]

Coverage increased (+0.005%) to 96.169% when pulling 81c40e3 on delay_waterfall into ac062da on master.

[nkern]

Thanks for getting a start on this, its much-needed! One comment is, I'm not sure if its right to plot a single waterfall containing multiple baseline-pairs. It seems a little odd to do that, and perhaps is just confusing. Seems like when we say "waterfall", we should mean a single baseline-pair with a non-zero time axis which makes it a heatmap. If we want to make multiple blpair heatmaps, instead of concatenating them into a single heatmap, we should be making multiple subplots... I can make this edit if you are ok w/ it and push to the branch

[philbull]

The idea was that this could eventually be used to make wedge plots too, which is why I allowed it to take multiple blpairs as an argument. It's not quite there yet though (e.g. it doesn't do the necessary averaging for you), so I agree more work is needed.

[nkern]

Okay, I pushed some edits that, if you aren't a fan of, we can revert. Here is the before & after

For a uvp with three baseline pairs and 3 time integrations per blp, before is

after is

For me, the importance is the distinction between different blpairs having their own subplot, otherwise it gets hard for me to interpret exactly where one ends and the other begins.

The other important changes are:

1. the units say log_10 P(k), but the colorbar reads 10^12, 10^13, etc which I think is confusing: if the units are log_10 P(k) the the tick labels should be 12, 13, etc and if its just P(k) then the units should be 10^12, 10^13, etc.

2. y-axis labeling, which is a little tricky b/c we want to label them in LST which wraps, so you can't just do ax.matshow(..., extent=[dly.min(), dly.max(), lst.min(), lst.max()], but I added some code to do the gymnastics for you

[nkern]

This stuff looks good to me. Phil let me know if you have any comments on the edits. I rebased the branch of master already, so should be good to merge once you've had a look.

Also about your comment on wedge plots: I think wedge plots are fundamentally different though. We can't make a "wedge-waterfall" because you've already used up your two heatmap axes for k_perp and k_parallel. A "waterfall" to me implies a time and spectral axes, so if we want to plot wedges, we should probably just make a new function (delay_wedge) to do that.

[acliu]

Is there an option to allow all the subplots to be on a common colour scale?

[nkern]

@acliu yes there is. you can specify the vmin and vmax parameters and all the colorscales are set to those limits

[nkern]

@philbull the other important change is that I enforced matplotlib>=2.2 in the setup.py b/c I had matplotlib=2.0 and it failed due to the cbar.get_yticks call (even though I see you had an explicit catch for it, it didn't work for me)... as well as numpy>=1.14 b/c that was causing some problems for me locally due to a numpy==1.13 installation...