[gloas] fork-choice compute_head performance issue

### Describe the bug

this is what happened on a network of <2000 validators (epbs-devnet-0)

this is way higher than mainnet (>2M validators)

### Expected behavior

it should take way less than mainnet

### Steps to reproduce

_No response_

### Additional context

`computeDeltas()` is used to be the main performance bottle neck on mainnet, it's great in `epbs-devnet-0`
so the issue should be somewhere during `this.protoArray.applyScoreChanges()`

### Operating system

Linux

### Lodestar version or commit hash

unstable/epbs-devnet-0