Add support for unpacking 16-bit BGRA

[gmarkall]

Adds support for unpacking 16-bit BGRA format. This format is used by ImageMagick 6 for many platforms - the addition of this support enables converting directly from Magick::PixelPacket* into PIL images.

I have used this new functionality in an example image-processing library I wrote to accompany a talk on Python APIs for CUDA-accelerated applications (see e.g. https://github.com/gmarkall/numba-accelerated-udfs/blob/main/filigree/api.py#L47), but I think it is generally useful for other applications to be able to work directly with ImageMagick 6 data.

[nulano]

Related to #1888.

It seems this PR scales 16-bit images to 8-bit by discarding the lower 8 bits. Note that this was considered controversial for I mode images in #3011.

[gmarkall]

It seems this PR scales 16-bit images to 8-bit by discarding the lower 8 bits.

Yes, it does - I just followed the same pattern used in unpackRGBA16L and unpackRGBA16B which seem to do the same:

Pillow/src/libImaging/Unpack.c

Lines 1010 to 1030 in dae82d1

	void
	unpackRGBA16L(UINT8 *_out, const UINT8 *in, int pixels) {
	int i;
	/* 16-bit RGBA, little-endian order */
	for (i = 0; i < pixels; i++, _out += 4) {
	UINT32 iv = MAKE_UINT32(in[1], in[3], in[5], in[7]);
	memcpy(_out, &iv, sizeof(iv));
	in += 8;
	}
	}
	void
	unpackRGBA16B(UINT8 *_out, const UINT8 *in, int pixels) {
	int i;
	/* 16-bit RGBA, big-endian order */
	for (i = 0; i < pixels; i++, _out += 4) {
	UINT32 iv = MAKE_UINT32(in[0], in[2], in[4], in[6]);
	memcpy(_out, &iv, sizeof(iv));
	in += 8;
	}
	}

(in fact I now realise I have a copy/paste error in the comment from those functions - will fix shortly)

[gmarkall]

(in fact I now realise I have a copy/paste error in the comment from those functions - will fix shortly)

... actually I didn't - apologies!... I found the comments I added confusing, but they're consistent with the other comments in the file as far as I can tell.