storage: improve block load metrics

We currently have these two metrics:
```
metaBlockBytes = metric.Metadata{
Name: "storage.iterator.block-load.bytes",
Help: "Bytes loaded by storage engine iterators (possibly cached). See storage.AggregatedIteratorStats for details.",
Measurement: "Bytes",
Unit: metric.Unit_BYTES,
}
metaBlockBytesInCache = metric.Metadata{
Name: "storage.iterator.block-load.cached-bytes",
Help: "Bytes loaded by storage engine iterators from the block cache. See storage.AggregatedIteratorStats for details.",
Measurement: "Bytes",
Unit: metric.Unit_BYTES,
}
```

In practice, the most likely figure we would be interested in is the difference between these two, i.e. how many bytes we read. We should replace the total value with the non-cached value.

Jira issue: CRDB-39220