texture atlas builder: use correct pixel size instead of 4

[mockersf]

Objective

• Fixes TextureAtlasBuilder::finish(), produces incorrect initial pixel byte data for texture formats with more then 1 byte per channel #2919
• initial pixel was hard coded and not dependent on texture format

Solution

• Replace the hard coded pixel with one using the texture pixel size

[DJMcNab]

Obviously it's possible to avoid this allocation, but it might need to change the interface of new_fill

Alternatively, we could just have a static array of zeroes long enough and just slice into it.

[mockersf]

Obviously it's possible to avoid this allocation, but it might need to change the interface of new_fill

new_fill is able to take more than one pixel and repeat it across the texture. It seems hard to modify it for this without losing this functionality... but then it's something that was done a long time ago 6164ea6 and I'm not sure anyone used it that way or is even aware of that...

Alternatively, we could just have a static array of zeroes long enough and just slice into it.

Today the max size is 16, but I suppose that could change and this would be forgotten.

[DJMcNab]

We could add a comment to the pixel_size function saying to check this method, or else make the pixel size a macro and have a way to get the largest one as a constant

[DJMcNab]

We could add a comment to the pixel_size function saying to check this method, or else make the pixel size a macro and have a way to get the largest one as a constant

[mockersf]

using the new method instead of new_fill, now we're back to only one alloc 👍

[DJMcNab]

Awesome, thanks!

[mockersf]

replaced by #2977 on pipelined rendering branch