admission: investigate read bandwidth spikes in stable workload

While running `admission-control/disk-bandwidth-limiter`, we notice that sometimes read bandwidth has outsized spikes in a seemingly stable workload. We should try to figure out what is causing these spikes. The leading theory is that we are seeing an uptick of both block and page cache miss rate which is forcing disk reads. But we should verify that using pebble metrics.

A better understanding here should help influence the design of https://github.com/cockroachdb/cockroach/issues/107623, where we get a better understanding of how reads impact the bandwidth/IOPS of the system.

Jira issue: CRDB-44327

Epic CRDB-42949