HealpixPixel supports numpy arrays as inputs

Currently we use a single `HealpixPixel` object to represent one pixel, which means for multiple pixels we end up creating lists of objects. This is relatively slow, and there are multiple points in the code where we have to convert between a list of `HealpixPixels` to numpy array of orders and pixels. It would be good to support a single `HealpixPixel` object containing numpy arrays of orders and pixels to make this conversion a lot faster, and we can replace Lists of `HealpixPixel`s with a single object with arrays.

Start by locating the HealpixPixel class and the code that converts lists of HealpixPixels to NumPy arrays of orders and pixels. Trace how single-pixel and multi-pixel inputs are used before defining the array-backed representation. Done means one HealpixPixel object can contain NumPy arrays and the relevant list-based conversions are replaced without changing behavior.

Written by the indexing model from the issue text.