Implement a performant bucketing MSM

[winston-h-zhang]

Works towards #65. This PR implements a serial version of a bucketing MSM algorithm, and reduces gas costs for large MSMs. We have the following tests on N=8,16, where improvement can already be seen:

Running 5 tests for test/multiscalar-mul-tests.t.sol:MultiscalarMulTest
[PASS] testGetAt() (gas: 3832)
[PASS] testGrumpkin3Serial() (gas: 22090653)
[PASS] testGrumpkin3Unoptimized() (gas: 22876388)
[PASS] testGrumpkin4Serial() (gas: 29651565)
[PASS] testGrumpkin4Unoptimized() (gas: 46444653)

Todo:

• Signed-digit optimizations
• Look into optimizing ff/ec routines

[winston-h-zhang]

I'm just now reading the comments in #60.

@storojs72 what would you think is the best way forward? I can continue working on some more optimizations, if that's something we need. We can also merge this as a somewhat self-contained proof-of-concept, and come back to it

[huitseeker]

@winston-h-zhang see also #65 (comment): I think we can merge this for now and iterate afterwards.