[Solver] Dancing Links Implementation

[mbrown1413]

Make assembler faster by implementing dancing links exact cover algorithm.

• Wikipedia
• Knuth's paper and book
• Consider "secondary constraints" addition to cover problem, like implemented in BurrTools and github.com/Seemee/exactcover. Columns which can have at most one 1 (but may have zero 1s) in the solution set of rows.
• Existing solvers to reference: burrtools, circular_assembler

After implementing, there are still plenty of other options for a faster solver covered in #45

[mbrown1413]

Implemented in:

• ff3df446cc8ba6bdceaff676a4d8808bc3a8d66a - Exact cover solver
• 62433e2630c6d1a1c558b67933c2752fd27cb052 - CoverSolver column range support
• 2de3baccef017f0a001d9364845df4bc445beaf8 - Change assembler to use CoverSolver
• af105b4d388a049205bc9b235e43bc8a72bd30e0 - Fix CoverSolver header corruption bug
• 3ea0e780b1df19055752189a587ece7d2ff1b093 - CoverSolver fixes and cleanup

I implemented a bit more than the standard dancing links algorithm, allowing for column ranges. This is what we use for optional voxels, piece instancing, and, in the future piece min/max ranges.

I haven't implemented progress or really tried hard to profile and optimize the algorithm like I originally planned. When I briefly tried profiling, it turns out that the the process of enumerating piece placements is slower than the actual assembly! I'm sure this won't be true for every puzzle, but for now I think this is good enough.