fix: optimize polynomial folding process in batch FRI prover

[crStiv]

This PR optimizes the polynomial folding process in the batch FRI prover by performing folding directly in the value domain instead of the coefficient domain.

Technical details:

• For a polynomial P(x) = sum_{i<r} x^i * P_i(x^r), we compute sum_{i<r} beta^i * P_i(x) directly in the value domain
• This is mathematically equivalent to the original folding in coefficient domain
• While we still need one final IFFT transformation for compatibility with the rest of the proof system, we avoid intermediate FFT transformations
• The optimization reduces the number of field operations needed for folding

The change maintains the same mathematical properties while making the implementation more efficient.

Changes:

• Removed coefficient domain folding
• Implemented direct value domain folding with detailed documentation
• Added explanatory comments for better code maintainability

The optimization addresses the TODO comment in the original code that suggested this improvement.