perf(db): flatten HashedPostState before persisting

[gakonst]

Summary

Accumulate hashed state from all blocks using HashedPostStateSorted::merge_batch and write it in a single batch, reducing cursor open/close overhead from N calls to 1.

This mirrors the existing optimization for trie updates (which was already batched).

Changes

• Move write_hashed_state call outside the per-block loop
• Use HashedPostStateSorted::merge_batch to combine all hashed states (newest-to-oldest)
• Write the merged hashed state in a single operation

Closes #20609

[mediocregopher]

## Commits
- **Baseline**: afe164b (main)
- **Feature**: 76a8ed0 (flatten-hashed-post-state)

## Results

| Metric | Baseline | Feature | Change |
|--------|----------|---------|--------|
| Mean Latency | 37.89ms | 37.36ms | -1.39% ≈ |
| Std Dev | 24.23ms | 23.98ms | |
| P50 Latency | 30.09ms | 29.76ms | -1.11% ≈ |
| P90 Latency | 69.32ms | 66.38ms | -4.23% |
| P99 Latency | 138.29ms | 135.17ms | -2.26% |
| Gas/sec | 792.98 Mgas/s | 804.13 Mgas/s | 1.41% ≈ |```

[yongkangc]

nice, didnt know that reducing cursor overhead would be so big