[witness gene] optimize StepRecord only copy once (#1250)

### design rationales
previous flow: tracer generate steprecord -> dispatch steprecord to
respective witness table
now: tracer generate steprecord -> dispatch steprecord index only ->
each table take step record index then fetch from steprecord

it improves the performance to avoid per record copy twice within memory
due to the ownership taken.


### benchmark
tested on `23817600` `generate_witness`

| Shard | master (s) | optimized (s) | Improvement (%) |
|-------|------------|----------------|------------------|
| 0     | 6.88       | 4.13           | 39.97%           |
| 1     | 3.97       | 2.83           | 28.72%           |
| 2     | 3.72       | 2.85           | 23.39%           |
| 3     | 3.66       | 2.83           | 22.68%           |
| 4     | 2.80       | 2.19           | 21.79%           |
| 5     | 2.37       | 1.99           | 16.03%           |
| 6     | 4.34       | 4.42           | -1.84%           |
| 7     | 6.40       | 3.78           | 40.94%           |
| 8     | 5.94       | 3.76           | 36.64%           |
| 9     | 5.69       | 3.83           | 32.69%           |
| 10    | 5.78       | 3.85           | 33.38%           |
| 11    | 3.80       | 2.72           | 28.42%           |
| 12    | 3.57       | 2.66           | 25.49%           |
| 13    | 3.51       | 2.59           | 26.21%           |

Author: hero78119

ceno_emul/src/lib.rs

@@ -10,9 +10,9 @@ pub use platform::{CENO_PLATFORM, Platform};
 
 mod tracer;
 pub use tracer::{
-    Change, FullTracer, LatestAccesses, MemOp, NextAccessPair, NextCycleAccess, PreflightTracer,
-    PreflightTracerConfig, ReadOp, ShardPlanBuilder, StepCellExtractor, StepRecord, Tracer,
-    WriteOp,
+    Change, FullTracer, FullTracerConfig, LatestAccesses, MemOp, NextAccessPair, NextCycleAccess,
+    PreflightTracer, PreflightTracerConfig, ReadOp, ShardPlanBuilder, StepCellExtractor, StepIndex,
+    StepRecord, Tracer, WriteOp,
 };
 
 mod vm_state;

ceno_emul/src/test_utils.rs

@@ -24,7 +24,8 @@ pub fn keccak_step() -> (StepRecord, Vec<Instruction>) {
         Default::default(),
     );
     let mut vm = VMState::new(CENO_PLATFORM.clone(), program.into());
-    let steps = vm.iter_until_halt().collect::<Result<Vec<_>>>().unwrap();
+    vm.iter_until_halt().collect::<Result<Vec<_>>>().unwrap();
+    let steps = vm.tracer().recorded_steps();
 
     (steps[2].clone(), instructions)
 }