feat(optimizePathOrder): new plugin

[KTibow]

Paths can often be optimized by changing their direction.
Base path: M16 14v-2H8v2Z (14b)
Optimized by changing the start: M8 14h8v-2H8Z (13b)
Optimized by reversing direction and changing the start: M8 12h8v2H8z (12b)
This plugin automatically does that!

Flow of the plugin:

• For each path, break it down into "parts". Each part is a continuous set of instructions. Parts are "valid" if they contain only instructions that we can move around and reverse properly.
• Send it off to be optimized
  • Remove the args, we'll recalculate them with coords. Any args that have special meaning go into a special property.
  • Disable changing the start if it has a non-round stroke, or if the part isn't closed.
  • Disable changing the direction if the fill rule is nonzero and there's more than one part in the path.
• Try with reversing (if available) and without reversing
  • Input: reverse the base and coords, and reverse the order if we're reversing this time
  • Output (mutable): an actual set of path commands.
  • Iterate through the possible orders
    • We essentially want to move the first command to the end
    • Remove the first command
    • Adjust the move command to move to the new first command's base, so that it isn't broken
    • Remove the previous last command and re-add it, except this time it can't be z
    • Add the new last command
    • Estimate the length of this new output. If it's better than anything so far, use it.
• Reconstruct the path from parts
  • If the optimization gets something better than the original, use that
  • If it doesn't or the part was "invalid", use the original

Closes #1847
Closes #78 (damn thats old)

This is disabled by default, due to performance worries. On single pass Isometric Madness, it usually takes around 100ms.

Results

Name	Standard	Multipass
Arch Linux logo	-95b	-102b
Blobs	-116b	-117b
Isometric Madness	-1637b	-2563b
tldr-pages banner	0b	0b
Wikipedia logo	-481b	-502b

* Upon merging the latest changes, I think something elsewhere changed making it less efficient, as the figures don't match up with above anymore.

[KTibow]

If I don't do this, whenever I set an exact value for an argument it leaks out, despite my plugin having a default set. (js2path stores the value exactly, and if it gets called later with a different precision it overrides.)

[anthonyryan1]

I just thought I'd chime in. This is the first plugin thing I've seen in years that could improve the compression of a few of my hand-optimized ultra small files.

I had previously thought there was little more optimization potential in them.

Great idea