Load test: chunk pre-fetch reconnect improvement

This is intended to test https://github.com/hiero-ledger/hiero-consensus-node/pull/25625

Measure the impact of chunk pre-fetch on reconnect duration with a large state.

### Setup

- 300M accounts
- Trigger reconnect after state is fully built
### Runs

| # | Branch | `virtualMap.chunkPrefetchDepth` | Purpose |
|---|--------|-------------------------------|---------|
| 1 | `main` | n/a (no pre-fetch code) | Baseline |
| 2 | `24636-reconnect-improvement` | `1` (default) | Safe option — expected moderate improvement |
| 3 | `24636-reconnect-improvement` | `-1` (unbounded) | Aggressive option — best possible gain, may increase memory |
| 4 | `24636-reconnect-improvement` | `0` (disabled) | Control — must match baseline |

### Metrics to capture

- Total reconnect wall-clock time (primary metric)
- `totalStallMs`, `prefetchExhaustedMs`, `maxLookahead` from the traversal summary log (INFO level, `RECONNECT` marker)
- Memory usage during reconnect (watch for elevated heap in run 3)
### Expected outcomes

- Run 4 ≈ Run 1 (confirms depth=0 is behaviorally identical to no pre-fetch)
- Run 2 ≤ Run 1 (pre-fetch overlaps stall time with useful work)
- Run 3 ≤ Run 2 (deeper lookahead may or may not help — `prefetchExhaustedMs` from run 2 tells us)