refactor rowmajormatrix witness

[hero78119]

This PR refactor mpcs api to take plonky3 rowmajormatrix as batch_commit input

in the preparation for next PR to support

• use plonly3 dft to encode RS code from lagrange domain
• use plonky3 mmcs to commit merkle tree

Example of new basefold.commit via plonky3 lib

in this case, we can skip monomial basis entirely by changing respective codeword fold in (1-alpha), (alpha)

// rmm is row major matrix witness in lagrange basis (evaluation basis)
let mut m = rmm.bit_reverse_rows().to_row_major_matrix(); // dft(reverse_row_bit(message)) == basefold::rs_encode(message)
// run fft, which is equivalent to interbolate binary field
m.pad_to_height(expansion_factor, E::Basefield::ZERO); 
let codeword = self.fft.dft_batch(m).to_row_major_matrix();

let (comm, _) = let mmcs.commit_matrix(codewords); // merkle tree commitment

change scopes

• wrapped plonly3 rowmajormatrix and retain ceno original rowmajormatrix padding functionality
• separate witness/structural_witness into different rowmajormatrix

[hero78119]

benchmark in fibonacci e2e with 2^20

fibonacci_max_steps_1048576/prove_fibonacci/fibonacci_max_steps_1048576
                        time:   [3.9887 s 4.0009 s 4.0147 s]
                        change: [-1.6824% -0.8643% -0.0106%] (p = 0.08 > 0.05)
                        No change in performance detected.

shows no impact on current refactor before/after, which is expacted