Clean up masking logic

[jeremykubica]

This PR refactors the masking code to clean up the interfaces, but does not change end-to-end behavior. A major part of this change is that most masking operations will now work on the LayeredImage's mask layer. For example, grow_mask will now expand the masked pixels in the mask layer instead of the science layer. LayeredImage's apply_mask() function will then apply the masks to the science and variance layers.

The motivations for these changes are:

• Support binary masks coming via the standardizers and WorkUnits. These will be cases where all the masking logic is applied when the WorkUnit is constructed.
• Maintain backwards compatibility with current input files.
• Consolidate most masking logic in LayeredImage instead of across RawImage, LayeredImage, and ImageStack.
• Reduce logic in RawImage for possible eventual removal of that class.

More detail on the changes:

LayeredImage: Move the majority of the masking logic into LayeredImage and perform most changes on the mask layer (instead of directly onto science and variance).

RawImage: Remove grow_mask function. All mask changes should be done through LayeredImage and then applied to RawImage with apply_mask().

ImageStack:

• Remove masking functions that just loop over the individual LayeredImages, such as apply_mask_flags().
• Remove global mask as an object attribute of ImageStack. Replace it with a function to create a new global mask on-demand. This allows the code to skip allocating a global mask in cases where no global masking is done.

masking.py:

• Remove the class-based logic. This was added to support use cases that will be covered in the standardizer. We only need to apply the individual functions from LayeredImage and ImageStack. The new flow is much simpler.

run_search.py:

• Move the masking logic into masking.py.

Closes #415 and #352.

[wilsonbb]

LGTM, just a few nits