Investigate memory usage in full Data Node mode

If you run Omnibus with all the data features turned on (see [this config](https://github.com/input-output-hk/acropolis/blob/prc/architecture/processes/omnibus/configs/ledger-with-api-and-history.toml), currently in prc/architecture only), it
climbs to 17GB by epoch 291 (I stopped it then) and outputs loads of Fjall errors - e.g.

```
2026-01-05T15:26:23.371085Z INFO fjall::partition: partition: write halt because of write buffer saturation
```

others (lost in scrollback) about delaying 64us because L0 cache full etc.

These might be two different things - see also #551 - but needs profiling.