Speed-ups

[jlhumber]

do some additional profiling to see where improvements can be made.

• [x] improved flattening
• [x] list comps become array operations
• [x] zips become column slices
• [x] chunking becomes reshape
• [x] packing
• [x] unpacking
• [x] buffer

add numpy and bottleneck dependencies