Trim empty space around edges of Shp(TD) frames

[PunkPun]

Revives #15148. It appears the original issues were fixed along the years. I suspect the addition of anti-aliasing had an effect on it.

#15184 (comment) is still present but we can't do anything about it. The 2nd option was chosen back then and we can choose it again. Auto bounds is correct much more often than its not anyhow.

[PunkPun]

The number of sheets generated by --dump-sequence-sheets go down by

TD 2 -> 1
RA 2 -> 1

[RoosterDragon]

Could we get a list of items where the selection bounds have changed? Could be useful for any follow-up that wants to tidy those up.

[PunkPun]

This is the diff e4be18d

Generated with 69fc1e1

[pchote]

#15194 was implemented as a much more robust solution to batch breakage - meaning that packing everything into one texture was no longer needed.

Do we have any perf numbers to indicate that trimming is worthwhile under this newer system?

[PunkPun]

as the sprite sheets are uncompressed, the empty space is eating up quite a bit of RAM. So removing empty sheets definitely way more impact than all the micro-optimisations we do

[pchote]

Only 16MB at the default sheet size, so i'm unconvinced that its worth the potential disruption (and that has definitely been true for most of those other optimisations...)

[RoosterDragon]

Yes the focus now would just be on the memory and not the batching gains. <100 lines to reduce our memory footprint by 5% is a nice win in my opinion. There was some pain when it was added originally, but since we've already dealt with the pain of finding and fixing the regressions, we've already paid those costs.

[abcdefg30]

Code changes lgtm. It seems like this is improving the bounds compared to bleed, e.g. with the badger or civilian houses.
Bleed:

This PR:

Although the latter could still be improved, I didn't spot anything bad on a quick check in RA.

[abcdefg30]

Changelog