Calculate complementary polynomial for QSP using flip convolution.

[Epsilon1024]

Calculate complementary polynomial using Eq. 60 of the paper

[Epsilon1024]

There are two small followups I will be adding after this PR is submitted:

1. This version does not normalize P before calculating Q. (From my understanding the first version of the default QSP did not normalize P, but this function was added later). I will be adding the normalization in a follow-up PR (that is nearly finished).

2. The main challenge that I faced in this PR was setting a high tolerance for the case with all real polynomials. I found the method in this PR performs comparatively to that in the code provided by the authors of the paper. I will write out my findings in a notebook to go along with this.

[anurudhp]

Looks good! Mostly docstring+type-hint nits. Also please fix the rng comment, best practice is to use np.random.Generator.

[Epsilon1024]

As a note: I ran some checks to see if normalizing P before running the tests on fast_complementary_polynomial would give us more accurate results. From my checks, it performs slightly worse than not being normalized. This tells me that the precision problem is not related to normalization.

[tanujkhattar]

@anurudhp @Epsilon1024 Can we merge this and continue improvements in follow-up PRs? Maybe open an issue to track the potential improvements?

[anurudhp]

@tanujkhattar good to go % hardcoded random seed

[anurudhp]

LGTM!