Change pixel size of "LA", "La", and "PA" from 4 to 2

These image modes currently use 4 bytes for each pixel, but they only need two. I've started looking into this, but there are a lot more changes than I expected, so I'm creating this issue for documentation. The files I've found that need to be changed are:

• /src/libImaging/Access.c
  • ImagingAccessInit() contains a mapping to three functions (also defined in this file) for each of these modes.
• /src/libImaging/Bands.c
  • Lots of special casing for "LXXA", and some functions assume an image with 2 bands uses 4 bytes per pixel.
• /src/libImaging/Convert.c
  • All of the conversion functions for these modes.
• /src/libImaging/Geometry.c
  • bicubic_filter32LA() and bilinear_filter32LA().
• /src/libImaging/GetBBox.c
  • ImagingGetBBox() creates a bit mask for the alpha channel.
• /src/libImaging/Imaging.h
  • IMAGING_PIXEL_LA and IMAGING_PIXEL_PA are defined in this file, though they aren't used anywhere.
• /src/libImaging/Jpeg2KDecode.c
  • j2ku_graya_la() is currently used for both "LA" and "RGBA", so it will first have to be duplicated and renamed for "RGBA", and then the original can be modified.
• /src/libImaging/Jpeg2KEncode.c
  • j2k_pack_la().
• /src/libImaging/Pack.c
  • The references to packLA() can be replaced with copy2(), and packLAL() needs to be modified. packLA() is no longer used after this change, so it can be removed.
• /src/libImaging/Paste.c
  • fill_mask_L() uses a mask for the alpha channel, and ImagingPaste() calls the same function for "LA" and "RGBA".
• /src/libImaging/Storage.c
  • The image pixel size and line size are set in ImagingNewPrologueSubtype().
• /src/libImaging/Unpack.c
  • unpackLA() and unpackLAL().
• /src/PIL/PyAccess.py
  • The mode_map uses _PyAccess32_2() for these modes.

Changing im->pixelsize from 4 to 2 also means the image data will be stored in im->image8 instead of im->image32. im->image also exists and always holds the image data. The only difference between these pointers is their type: image is char, image8 is UINT8, and image32 is INT32. Some functions check for im->image8 being set and do different things based on that.

Start by tracing ImagingNewPrologueSubtype() in src/libImaging/Storage.c and ImagingAccessInit() in src/libImaging/Access.c, then follow the listed LA, La, and PA handling across the C files and src/PIL/PyAccess.py. Confirm that the modes use a two-byte pixel size, image8 storage, and correct access, conversion, geometry, JPEG 2000, packing, pasting, and unpacking behavior.