db: better testing of error handling pathways

cockroachdb/cockroach#63847 is likely an example of improper error handling introducing corruption. As a part of the quality initiative, we should make an effort to audit and improve testing of error handling code paths.

We currently have some existing tests like [TestIngestError](https://github.com/cockroachdb/pebble/blob/108cd0260986c5972d20bd807a4d7192d912c97a/ingest_test.go#L581) that try injecting an error at every _i_-th operation. There are probably more code paths that deserve this testing treatment. That particular test, `TestIngestError`, might also benefit from more assertions around the state of the resulting DB.

The metamorphic test has an [`-error-rate`](https://github.com/cockroachdb/pebble/blob/master/internal/metamorphic/meta_test.go#L53-L54) that will inject errors into _read-only_ operations and automatically retry them, but it currently defaults off. We can update it to default to some low rate.

Related to #270.

Jira issue: PEBBLE-212

Start with TestIngestError in ingest_test.go and the -error-rate handling in internal/metamorphic/meta_test.go. Review how these tests inject errors and what database state they assert, then identify additional error-handling paths that need similar coverage. Done means stronger state assertions and a low default error rate for the metamorphic test.

Written by the indexing model from the issue text.