Ensure Image and core palettes are in sync after PA conversion

[akx]

While working on #9833, I noticed that an RGBA palette's alphas would be "resurrected" if you converted from PA to P (which retains the RGBA palette in the core im object even if the ImagePalette wrapper is recreated), and then to RGBA.

With this, the "zombie" alpha from the RGBA doesn't appear anymore.
I would suspect this is more the Do What I Mean behavior, because an RGBA palette's alpha values weren't "visible" in a PA image anyway.

[radarhere]

Suggested change

	im.putalpha(Image.frombytes("L", (2, 1), bytes([240, 220])))
	assert im.get_flattened_data() == ((0, 240), (1, 220)) # Matches the alpha band

These lines don't appear relevant.

[akx]

They're meant to show the palette alpha isn't used.

[radarhere]

I consider it pretty foundational that the palette doesn't affect pixel values (without conversion). I'm not sure it needs to be tested here? We already have

Pillow/Tests/test_image_putalpha.py

Lines 28 to 33 in abb8ba0

	im = Image.new("P", (1, 1), 1)
	assert im.getpixel((0, 0)) == 1
	im.putalpha(2)
	assert im.mode == "PA"
	assert im.getpixel((0, 0)) == (1, 2)

[akx]

That test_image_putalpha segment doesn't seem to really cover the contents of the palette.

I don't see how having this here really hurts?

[radarhere]

Suggested change

	from . import ImagePalette
	new_im.palette = ImagePalette.ImagePalette("RGB", im.getpalette("RGB"))
	new_im.palette.dirty = True
	new_im.load()
	new_im.putpalette(im.getpalette())

[akx]

Applied but with explicit "RGB" + comment.

[radarhere]

Out of curiosity, why explicit "RGB"?

[akx]

Because then you don't need to look up getpalette()'s definition to find out that the default is in fact "RGB", and not something that's inferred from im, as I would've initially assumed.

(Or, IOW, Explicit Is Better Than Implicit.)