Plonky2 prover in Cuda

[doutv]

Implement all these code in Cuda, reduce CPU <-> GPU communication, data transfer between CPU and GPU only happens once.

Starting from here: "commit to quotient polys", PolynomialBatch::<F, C, D>::from_coeffs
All data should remain in GPU. After finishing proving, transfer the final proof to CPU

https://github.com/okx/plonky2/blob/b74c2ac48bd25df0cbdee2e4b3871f63b631178d/plonky2/src/plonk/prover.rs#L281-L337

This is a meta issue

[doutv]

• "commit to quotient polys" already implemented in cuda. https://github.com/okx/plonky2/blob/b74c2ac48bd25df0cbdee2e4b3871f63b631178d/plonky2/src/plonk/prover.rs#L281-L292
• "construct the opening set" Construct opening set #67
• "challenger" challenger #69
• Prepare opening for FRI, Compute final_poly and its lde Prepare opening for FRI #70
• FRI proof of work FRI proof of work #71
• FRI commit phase FRI commit phase #72
• FRI query phase FRI query phase #73

[doutv]

We can write unit test first, ensure the output of Plonky2 CPU code is the same as our GPU code. TDD method