chore: compute beta products for max log dyadic size instead of 21

[johnathan79717]

From running BB_BENCH=1 ./build/bin/client_ivc_bench --benchmark_filter='.*Full/5.*', we can see that the compute_combiner only takes up 53.3% of the time of combiner_quotient_round.

  ├─ ProtogalaxyProver_::combiner_quotient_round                                       [1]  20.4 %   1.18s      (107.14 ms x 11)
    ├─ compute_combiner                                                                  [2]  53.3 %   628.5ms    (57.14 ms x 11)
    └─ (other)                                                                           [2]  46.7 %   550.1ms   

After some investigation, about 40% of the time was spent on compute_beta_products to construct gate separators. Its size was a constant 2**21 but we only use it up to the dyadic size of the circuit. This PR passes the max dyadic size instead.

This change saved about 500ms from combiner_quotient_round in the Full/5 benchmark.

  └─ ProtogalaxyProver_::combiner_quotient_round                                       [1]  12.7 %   693.0ms    (63.00 ms x 11)
    ├─ compute_combiner                                                                  [2]  90.2 %   625.1ms    (56.83 ms x 11)
    └─ (other)                                                                           [2]  9.8  %

[iakovenkos]

lgtm!