db: collect effective read-amp metric

We maintain a read amplification metric that we surface in CockroachDB as a timeseries metric. This metric is the worst-case read amplification in terms of the number of sstables consulted.

We should collect additional metrics that allow us to measure the effective read-amp actually observed by readers too. Optimizations like #2366 (or REMIX or DISCO papers) do not reduce the worst-case read amplification but can significantly reduce the number of read I/Os in practice.

Jira issue: PEBBLE-528

Start by locating Pebble's existing worst-case read amplification metric and how it is surfaced to CockroachDB as a timeseries metric. Read the context in #2366 and compare the requested effective reader-observed read I/O measurements with the existing metric; done means additional useful metrics are collected and surfaced without replacing the worst-case measure.

Written by the indexing model from the issue text.