[Feature request] `rgbgfx -r` should encode the reverse image appropriately

[Rangi42]

Currently rgbgfx -r generates all its output images as 8-bit RGBA encoded. I would like if it tried to match the expected encoding of the original input image.

So, if there's no palette, it would output 2-bit grayscale (or 1-bit with -d 1). If there is a palette, it would be embedded in the output PNG as an indexed colormap. Also, there shouldn't be an alpha channel.

This would give smaller output filesizes. Also, it would be more likely to work with people's bespoke build processes (e.g. if they need grayscale or indexed input images, like pret does).

[pinobatch]

Would rgbgfx -r -c embedded trigger indexed output mode?

[Rangi42]

No, that would warn about no longer having the palette information anywhere, and output grayscale.

[Rangi42]

• No palettes, 2bpp: type = PNG_COLOR_TYPE_GRAY, depth = 2
• No palettes, 1bpp: type = PNG_COLOR_TYPE_GRAY, depth = 1
• 1-64 palettes, no alpha colors: type = PNG_COLOR_TYPE_PALETTE, depth = 8, PLTE chunk
• 1-64 palettes, some alpha colors: type = PNG_COLOR_TYPE_PALETTE, depth = 8, PLTE chunk, tRNS chunk
• >64 palettes: type = PNG_COLOR_TYPE_RGB_ALPHA, depth = 8