mpcs: cleanup bunch of unnecessary basefold bit reverse operation

[hero78119]

Extracted from #843
This PR retain nice property and clean up bunch of unnessesary bit reverse across mpcs basefold prover/verifier.
The core idea is to retain raw message in even-odd fold, which match the sumcheck implementation parameter binding order.

[hero78119]

For fibonacci 2^20 e2e it bring ~3-4% latency improvement

fibonacci_max_steps_1048576/prove_fibonacci/fibonacci_max_steps_1048576
                        time:   [3.8632 s 3.8857 s 3.9076 s]
                        change: [-4.6226% -3.7207% -2.7579%] (p = 0.00 < 0.05)
                        Performance has improved.