Add Resample Capabilities to NDCube

[DanRyanIrish]

Description

NDCube should have the ability to:

• Downsample the resolution of the cube via a number of different methods, e.g.:
  • mean
  • sum
  • min
  • max
  • median
• Collapse an axis, which is a special case of the first point
• Upsample based on N-D fits, e.g. spline fits.

The first pass of this functionality should only support upsampling or downsampling an integer multiple of the resolution along each axis. Later versions could support resampling onto a user-defined grid for each axis/axes. However, this may be beyond the minimalist scope of ndcube and perhaps more advanced resampling methods should be left to higher level subclasses of the ndcube data classes.

A possible API might be

my_cube.resample((1, 1/2, 3, 100), method=None)

where my_cube is a 4-D NDCube instance and the input is a tuple of the resample factor for each data axis, i.e. the size of each pixel as a factor of the original pixel size in that axis. In this scheme, 1 denotes no resampling.
A number, x where 0 < x < 1, e.g. 1/2, denotes the axis will be upsampled so that each pixel is half as big as the original pixel along that axis. In this case, x will be inverted and rounded to the nearest integer and that will be the upsample factor.
Numbers > 1, e.g. 3, denote that the new pixels will be 3 times as big as the original pixels along this axis. In the first pass, this number must be an integer factor of the number of pixels along that axis. Numbers > 1 will be rounded to the nearest integer.
A number that is >= the number of pixels along the axis collapses the axis entirely and reduces the dimensionality of the cube.
The method kwarg (or a better name) would denote how the resampling would be done. In cases where the resample factor is > 1, i.e. downsampling, the allowed values may be "mean", "sum", "min", "max", "median".
In cases where the resample factor is < 1, the allowed values may be "spline", or whatever fitting types are supported.

Additional context

[DanRyanIrish]

This an now be implemented using the new reproject functionality in astropy wcs.

[DanRyanIrish]

This issue is addressed by #450