Block matrix acceleration

[JamesYang007]

For any block matrix, we can make the following optimizations:

• Parallelize lasso-fitting on strong set for each block
• We only need to refit lasso for the blocks that increased in strong-set.

[JamesYang007]

• [x] Create a BASIL state class to manage all the internal variables (strong_set, active_set, etc.).
• [x] Refactor basil function using the state class. basil should be a really simple control-flow function.
• [x] Put lmda_max construction into another function that takes in a state object.
• [x] Generalize lambda sequence creation given lmda_max.
• [x] Add new block-matrix version of basil function. Think carefully about parallelization!