[perf] VerifyBatch direct hint

[nyunyunyunyu]

VerifyBatch chip hints merkle proofs directly instead reads from memory. This saves some cells for hinting and some columns of memory operations. This is valid because all merkle proofs are just read once.

Comparison on c7a.48xlarge: reth bench
Comparison on c8g.48xlarge: reth bench
Ecrecover leaf flamegraph before
Ecrecover leaf flamegraph after

Metric: used cells in Ecrecover leaf flamegraph
Before: 158,652,038
This PR: 142,929,350
Perf diff: -10%.

closes INT-3186

[jonathanpwang]

Please fill out Perf PR Template.

[jonathanpwang]

LGTM except update spec and many sibling_id were not renamed to proof_id

[github-actions]

group	app.proof_time_ms	app.cycles	app.cells_used	leaf.proof_time_ms	leaf.cycles	leaf.cells_used
verify_fibair	(-604 [-28.2%]) 1,536	(-308113 [-60.0%]) 205,675	(-8465997 [-45.2%]) 10,244,416	-	-	-
fibonacci_program	(-12 [-0.2%]) 4,988	1,500,095	51,485,080	(-795 [-11.5%]) 6,126	(-676416 [-35.1%]) 1,248,380	(-18587328 [-26.8%]) 50,830,327
regex_program	(+20 [+0.1%]) 14,813	1,914,103	165,455,373	(-2184 [-7.6%]) 26,586	(-1370872 [-23.3%]) 4,512,717	(-37670496 [-14.6%]) 221,226,909
fib_e2e	35,147	1,515,083	410,691,902	39,691	7,374,941	306,395,181
ecrecover_program	(-20 [-0.8%]) 2,514	284,567	15,055,723	(-1348 [-7.6%]) 16,470	(-636266 [-15.3%]) 3,521,596	(-17488962 [-9.4%]) 169,243,375

Commit: 8c0088c

Benchmark Workflow