fix(js): correct gas cost reporting by invoking step callback in step_end

[mattsse]

Summary

This fixes the long-standing issue where log.getCost() in JS tracers returned incorrect gas costs (paradigmxyz/reth#10959, #200).

Problem

The JS step callback was invoked in step() (before opcode execution), but the gas cost is only known after the opcode executes. This caused the first opcode to report cost=0 and each subsequent opcode to report the previous opcode's cost.

Before (incorrect): [0, 3, 3] for PUSH1, PUSH1, STOP
After (correct): [3, 3, 0] for PUSH1, PUSH1, STOP

Solution

The key insight is that we need the pre-execution stack/memory state but the post-execution gas cost. This requires a two-phase approach:

1. step(): Capture pre-execution snapshot (stack clone, memory snapshot, pc, op, etc.)
2. step_end(): Compute gas delta and invoke JS callback with the snapshot

Changes

• Add PendingStep struct to capture pre-execution state in step()
• Add MemorySnapshot to properly snapshot memory contents (SharedMemory clone is shallow due to Rc)
• Invoke the JS step callback in step_end() with captured state and correct gas cost
• Update tests to expect correct gas costs

Testing

Verified against live node that the fix produces correct gas costs:

curl -X POST http://node:8545 -d '{
  "method": "debug_traceTransaction",
  "params": ["0x...", {"tracer": "{ ... log.getCost() ... }"}]
}'

All existing tests pass.

Breaking Change Note

JS tracer errors that occur on the final opcode (like STOP) can no longer revert the transaction since the opcode has already completed. The transaction succeeds but the trace may be incomplete. This matches the expected behavior since the tracer is observational.

Fixes: paradigmxyz/reth#10959
Closes: #200

[yongkangc]

This is an interesting approach, makes sense