chore: try add small optimizations to divide_polynomialcoeff

[kevaundray]

This adds three optimizations:

• Replace .insert(0,x) with .append(x) and then a .reverse. Inserting at the first index on each loop causes every item in the list to be shifted down by one at each loop iteration.

• return [x % BLS_MODULUS for x in o] to return o . The values of o should already be reduced since they are coming from the div function.

• (a - b + modulus) becomes (a + (modulus - b)) - This mainly a nit, as the order of operations is not entirely clear.

[kevaundray]

This shaves off around 2 minutes from the 7594 tests (looked at other PRs to see how long they took compared to this)

[asn-d6]

This looks correct to me!

I'm approving but not merging, in case we want to discuss this further. Also curious how much each of the three (two?) optimizations contribute to the overall performance improvement: "returning o" is a clear improvement in performance and readability, whereas the append/reverse is slightly non-trivial (still fine though).

[kevaundray]

Good points -- I'm happy to leave this open until there's consensus