General PICT improvements

[andrews05]

This is an output size optimisation for pict encoding that automatically switches to indexed mode when there are no more than 256 colours in the image. It is a little slower though (even for images with too many colours, mainly due to the clut search in set), so not sure if it should be opt-in? (Or if you're happy to have this in Graphite at all?)

There are some other improvements here though, including some from the QuickTime PR.

• Added accessible format property to pict class, which stores the bit-depth.
• Reverted clut to using a vector again, padding out the size when the values aren't sequential. This restores the high performance of the original implementation for 'device' cluts while being just as fast for 'pixmap' cluts.
• Fast packbits decode for single byte runs. This results in around 10% faster PICT decoding.
• Combined the packbits encode functions as a template, and made some tweaks to be more willing to encode pairs as runs (improves the compression effectiveness a little).
• Fixed a rounding error in the 16-bit colour decoding I added earlier. It now accurately matches QuickDraw output.
• Removed encoding of the def_highlight opcode. I found this to be absent on many existing picts so I don't think it serves much purpose.

[andrews05]

I discovered that Nova doesn't scale paletted images nicely (e.g. for shipyard thumbnails) so I removed the automatic indirect encoding. I've left the function in but it's currently unused. There's a bunch of other helpful fixes and improvements though, so would still be good to merge this.