Prune too many checkpoint states at restart

### Describe the bug

@nflaig noticed this when he restarted his mainnet node

```
info: Removed persisted checkpoint states from the last run count=178, maxEpochsInMemory=3
```

also @philknows noticed in another mainnet node

```
Feb-09 03:32:48.623[chain] info: Removed persisted checkpoint states from the last run count=261, maxEpochsInMemory=3
```

### Expected behavior

with mainnet condition, it should be <= 3

### Steps to reproduce

_No response_

### Additional context

_No response_

### Operating system

Linux

### Lodestar version or commit hash

v1.27.0

Start by tracing the restart path that loads and prunes persisted checkpoint states, using the reported mainnet log messages as the symptom to investigate. Reproduce or inspect the pruning behavior under mainnet conditions and verify that no more than three checkpoint states remain in memory after restart.

Written by the indexing model from the issue text.