Feat: use openvm gpu prover for recursion

[kunxian-xia]

Summary

We added a new feature called gpu. Running with the cmd

RUST_LOG=info cargo test --release --package ceno_recursion  --lib  aggregation::tests::test_single  --features "gpu" -- --exact

We can see the gpu prover's performance is like this (876160 instructions).

execute_metered [ 20.6ms | 100.00% ]
execute_preflight [ 126ms | 66.19% / 100.00% ]
trace_gen [ 431ms | 0.93% / 100.00% ] (⬆️ 😬)
stark_prove_excluding_trace [ 183ms | 0.07% / 100.00% ]

And the cpu's prover print out log like this.

execute_metered [ 20.4ms | 100.00% ]
execute_preflight [ 115ms | 65.98% / 100.00% ]
trace_gen [ 252ms | 1.16% / 100.00% ]
stark_prove_excluding_trace [ 3.52s | 0.12% / 100.00% ]

aggregate binary

with gpu, we have

Aggregation - Start leaf proof (idx: 0) at: 1.031152699s
 INFO total_proof:stark_prove_excluding_trace: openvm_stark_backend::prover::coordinator: num_air=16
Aggregation - Completed leaf proof (idx: 0) at: 1.587258683s
Aggregation - Start leaf proof (idx: 1) at: 1.587274907s
 INFO total_proof:stark_prove_excluding_trace: openvm_stark_backend::prover::coordinator: num_air=16
Aggregation - Completed leaf proof (idx: 1) at: 2.065737664s
Aggregation - Start leaf proof (idx: 2) at: 2.065757212s
 INFO total_proof:stark_prove_excluding_trace: openvm_stark_backend::prover::coordinator: num_air=17
Aggregation - Completed leaf proof (idx: 2) at: 2.292657888s

without gpu, we end up with

Aggregation - Start leaf proof (idx: 0) at: 952.493241ms
 INFO total_proof:stark_prove_excluding_trace: openvm_stark_backend::prover::coordinator: num_air=16
Aggregation - Completed leaf proof (idx: 0) at: 7.603724584s
Aggregation - Start leaf proof (idx: 1) at: 7.603750731s
 INFO total_proof:stark_prove_excluding_trace: openvm_stark_backend::prover::coordinator: num_air=16
Aggregation - Completed leaf proof (idx: 1) at: 14.088209165s
Aggregation - Start leaf proof (idx: 2) at: 14.088250958s
 INFO total_proof:stark_prove_excluding_trace: openvm_stark_backend::prover::coordinator: num_air=17
Aggregation - Completed leaf proof (idx: 2) at: 18.048298693s