Feat/update compute_h function

[Fantoni0]

This PR updates the compute_h and compute_h2 functions in the pointvsline.rs file.
These functions iteratively fold multiple polynomials until a single one is computed as result.
The original implementation used DensePolynomial for the inner computations, which is fine. But for large MLEs this causes excessive allocations and intermediate objects that are soon discarded.

This PR makes 2 significant changes:

• Adds parallel iterators.
• Instead of using the DensePolynomial struct, we work over the coefficients of the polynomials directly.

This approach seems to be fine, I have compared the results against the original implementation (you can do this by replacing the compute_h function with the original implementation). I have also extended the tests cases for the test_compute_h2_compare test function.

Results:

• Parallel iterators do not help, might even slighltly worsen the results, for small MLEs. But I believe all our applications leverage large polynomials.
• Similarly, the optimizations do not seem to be significan for small MLEs. But for large cases, and successive foldings, it gets x2 or even x10 faster. You can run cargo bench --bench mova_matrix to test it. I tried size 16 and 512.

[NiDimi]

The changes look good. Tested the benchmarks and there is a noticable increase in perforamance when we fold multiple times. Great work Antonio 🔥. Can you add in the benches description the size of the matrices? Then we can merge it.