Marlin refactoring

[lgiussan]

This pull request addresses Issue #27. The high level changes are the following:

• There is no longer the need to explicitly pad R1CS matrices with dummy variables or constraints to make them square. Instead the padding is implicit in matrix-vector operations.
• The UnnormalizedBivariateLagrangePoly is replaced with the normalized LagrangeKernel.
• Matrix arithmetization is updated to use LagrangeKernel.
• Prover no longer needs to supply an oracle for T_eta(alpha, X).
• The query of the verifier to the oracle for T_eta(alpha, X) during outer sumcheck is substituted with the computation of the unique value v which satisfies the outer sumcheck. Later v is used in place of T_eta(alpha, beta) in the inner sumcheck. Explicit checking of the outer sumcheck equality is no longer needed with this optimization.
• The names of some variables, functions, and files have been changed to better reflect the notation of https://arxiv.org/abs/2107.04315

[DanieleDiBenedetto]

Okey I left unresolved the comment related to stuff that should be addressed in the monorepo. When the monorepo branch is stable let's bring the current Marlin inside it. Then we will open a PR addressing the remaining changes so we can keep track of them PS. The ToBytes deprecation was already performed by me in ginger-lib's sponge branch, so no need to address it here anymore.

[lgiussan]

Besides the requested changes, I did an additional change. At matrix construction time I re-index the columns of the matrix to be coherent with the treatment of public input domain X as a subdomain of the domain H. In this way there is no need to re-index matrix columns during computations.