storage: surface persistent storage errors

This issue tracks improving the observability of the system when Pebble is hitting recurring errors (especially IO errors).

This is inspired by a customer issue where compactions were erroring out with "too many open files". This lead to write stalls and eventually a crash due to an unrelated bug (triggered by the failures).

Some ideas:
- if Pebble is unable to complete flushes or compactions for some period of time, the node should be marked as "suspect" in the UI and there should be some kind of notice bubbled up.
- errors received from the filesystem should be observable e.g. in a metric

More ideas welcome @jbowens @sumeerbhola

Jira issue: CRDB-32064