WIP: Layers

[takluyver]

The idea of this is that you can 'overlay' raw & proc data and still be able to access either:

In [2]: run = open_run(2734, 182, data=('raw', 'proc'))

# Find the source in the first data section where it exists (here raw):
In [5]: run['SPB_DET_AGIPD1M-1/DET/10CH0:xtdf', 'image.data'].dtype
Out[5]: dtype('uint16')

# Get the source from a specific layer (syntax up for discussion):
In [7]: run.layer('raw')['SPB_DET_AGIPD1M-1/DET/10CH0:xtdf', 'image.data'].dtype
Out[7]: dtype('uint16')

In [8]: run.layer('proc')['SPB_DET_AGIPD1M-1/DET/10CH0:xtdf', 'image.data'].dtype
Out[8]: dtype('<f4')

I'm not sure about this yet - it makes bits of the code rather more complex, and it might not be used that often.

The aim is that you can have a combined view of the data - e.g. to do .select(..., require_all=True) - and specifically access e.g. AGIPD proc data, telling it that you want proc data so it fails if that's missing rather than quietly switching to raw (#292). We could even make data=('raw', 'proc') the default without changing things that already work (mostly).

@philsmt this would be an alternative to your #298. In this PR, the first named origin containing a source 'wins', as opposed to the last in your PR.

[takluyver]

It looks like we're going to move forwards with making new source names for corrected data, in which case this approach is unnecessary. We'll press ahead with #298 & #468 to handle that.