Implementation of resampling using reproject

[adwaitbhope]

Description

Expanding on #155.

NDCube should introduce a wrapper around reproject to implement resampling. As an initial implementation, the interpolation algorithm should be supported and the wrapper should be very thin. Later on, some of the parameters can be computed from data within the NDCube for simplicity.

This should apply to NDCubeSequence such that all the cubes are reprojected to the WCS of the first cube in the sequence, to get rid of the minor differences in the WCS of all cubes. The final WCS will need to be modified to include an extra dimension corresponding to the sequence.

The next step to follow could be trying to detect the invariant axes while resampling an NDCube. This would pave the way to perform resampling efficiently by reducing memory usage. It'll also help implement reproject's adaptive algorithm since it currently only supports resampling 2D arrays representing celestial axes.

Handling uncertainty and mask of an NDCube should also be taken care of, but it is not clear at the moment.

[DanRyanIrish]

Hi @adwaitbhope. This looks like a good description of what we discussed. To help outline and track the work, I've created a GSOC Resampling project and broken down the initial implementation into a number of issues. If you reference the correct issue in the description of your PR, it should be automatically added to the project.

As you progress we can add more issues to this project to lay out the next steps in more detail. For now, I've broken the initial implementation of the NDCube.resample method into 2 issues, #428 and #429, and captured the use of that method by NDCubeSequence as #430.