optimize parallel version sumcheck

[hero78119]

Related to #739

[hero78119]

naive parallel

sumcheck_nv_26/prove_sumcheck/sumcheck_nv_26
                        time:   [3.1378 s 3.1582 s 3.1823 s]

parallel with grid-tride-loop

sumcheck_nv_26/prove_sumcheck/sumcheck_nv_26
                        time:   [646.08 ms 651.60 ms 658.02 ms]
                        change: [-79.612% -79.368% -79.120%] (p = 0.00 < 0.05)
                        Performance has improved.

devirgo

devirgo_nv_26/prove_sumcheck/devirgo_nv_26
                        time:   [440.45 ms 442.60 ms 445.26 ms]
                        change: [-0.3165% +0.4289% +1.1818%] (p = 0.30 > 0.05)
                        No change in performance detected.

[hero78119]

New commit 31907d4 experiment bunch of unsafe api on fix_variable, however benchmark result shows slight slower

sumcheck_nv_26/prove_sumcheck/sumcheck_nv_26
                        time:   [1.8576 s 1.9913 s 2.1615 s]

Will investigate and look for different strategy