need to transfer surgery map between frames of a multiframe ncvisual

[dankamongmen]

When we run ncplayer -bpixel on a multiframe image (e.g. a video) when using Sixel (this does not happen in Kitty), the frame count and time get obliterated. We need to carry the surgery map through frames. We ought be able to embed it into the transparency-annihilation matrix as a third value. We can invoke this internally from ncvisual_decode() -- the user needn't deal with this at all.

[dankamongmen]

This is necessary for getting ncplayer reliably displaying the legend with Sixel (see #1432); things work well enough now with Kitty.

[dankamongmen]

To do this right, I think we need to hook the T-A matrix up to the ncvisual, either directly or through the sprixel. Let's think lifetimes here: the ncvisual must live through the last decode we want to do; the T-A must live through the last pixel blit we want to do....we need a T-A matrix whether it's multiframe media or not for sprixel drawing, but the ncvisual doesn't need it in that case. Maybe just comutual linkage, like we have between the sprixel and the ncplane?

[dankamongmen]

We have a handle to the ncplane within ncvisual_render(), and presumably we're reusing the ncplane--we have to, otherwise the sprixel would have been destroyed. and technically it's not a property of the ncvisual or sprixel, it's a property of the plane. If we blit some othet, unrelated bitmap to the same plane, we have the same cells atop it, so we can apply the T-A matrix unchanged (save layer 1, which must be recomputed each time we blit).

[dankamongmen]

we're reusing the tacache across renders now; we need to actually cut out the appropriate cells in the load.

[dankamongmen]

for the first time, we've got gaps maintained in the legend of ncplayer with sixel! =]

[dankamongmen]

yield no longer flickers in sixel =]