Fix Grayscale depth reduction with tRNS

[andrews05]

This is a fix for #426. The tRNS chunk was not being accounted for when reducing the bit depth of a grayscale image.
There are two possibilities we need to account for here:

1. If the tRNS value can be losslessly reduced to the new bit depth, we must do so.
2. If it can't be, this means the value does not appear in the IDAT and the tRNS must be removed.

I've added two tests for both of these cases.

The example images in #426 are in grayscale-8 with a tRNS value of 1 (near-black). This colour does not appear in the image and when the depth is reduced to 1-bit, the tRNS value of 1 now refers to white, making the image black and transparent instead of black and white.

Closes #426.

[edit]
Would be good if you could double-check my logic for validating the tRNS value. Or see if you have any ideas for achieving the same thing without a loop. I note the current grayscale reduction only supports reducing from 8-bit (not from 4 or 2), but the tRNS reduction/validation should continue to work even if that limitation is removed in future (it would be a bit of work though, so not inclined to do it right now).

[andrews05]

Just realised tRNS isn’t accounted for in 16 to 8 reductions either. It’s probably highly unlikely to happen but I should fix that too…

[andrews05]

So it seems the high bytes of the tRNS chunk are ignored for 8-bit, which means there's no need to reduce it during the 16 to 8 reduction.