Pack according to domains

[ma3ke]

Resolves #1

Requires some extensive testing, because we still want to see a nice and well-stirred result. Particularly for cases where the number of segments to be placed is smaller, we need to make sure that things are still uniformly distributed.

• [ ] Test whether the distribution is still nice and uniform where we need it to be.
• [ ] Is the quality of the packing still solid? (On Bart's suggestion, we can try to pack using the wave method proposed here and see whether we can still pack more structures onto that same space with our previous whole-space convolution strategy.)
• [ ] Investigate the performance hit.
• [ ] Investigate and describe the memory footprint improvement.

[ma3ke]

Just a quick in-progress pull request. Still working on it. See issue #1 for some info on what is left to do and left to verify before we merge this.