Move sr-arithmetic to a new crate and add in a fuzzer

[expenses]

This closes #3745. This commit splits up sr_arithmetic.rs into a new crate files, translates the previous 'fuzzing' tests into honggfuzz binaries (I was going off of parity-scale-codec which uses the same, happy to change fuzzing tool though) and adds a (currently failing) multiply_by_rational test that was uncovered via fuzzing.

@kianenigma

[parity-cla-bot]

It looks like @expenses signed our Contributor License Agreement. 👍

Many thanks,

Parity Technologies CLA Bot

[expenses]

The fuzzing code found that this line:

substrate/core/sr-primitives/src/sr_arithmetic.rs

Line 1667 in 0cd9bbf

if r > (c / 2) { q = q.add(&to_big_uint(1)); }

sometimes adds 1 to the value of the result when it shouldn't.

[kianenigma]

Looks good all in all! I'll take a last look before merging soon.

[bkchr]

We use https://crates.io/crates/criterion for benchmarking.

[expenses]

@bkchr I'm a little hesitant to make these changes to the arithmetic logic, because it might be better to save them for another PR. @kianenigma what do you think?

[kianenigma]

@expenses The recommendations don't seem to be any logic change but rather syntactic improvements. Those can be applied safely.

The migration to criterion can be saved for another PR as it is a bit bulky and I don't want to stall this PR for it too much. @bkchr I have added benchmarks in the past both with and without criterion. If we really want to have as a "we benchmark only with criterion", we should at some point a) make it a strict rule b) migrate the already existing ones.

Personally, I don't think it is worth migrating them and making it strict rule. Given that the required benchmark is fairly simple (like the ones in this crate) the test crate gets the job done okay.