Clarify and Update Meaning of Collection Table

This issue is an extension of https://github.com/rust-lang/rustc-perf/issues/937 which aims to add more clarity to the database schema.

This issue uses the new glossary for terms. Please make sure to first refer to that glossary.

Status Quo

Currently, the collection table is used to identify a single test.

For example, given an artifact, a benchmark "hello world", the profile "debug", and the "empty" incremental cache state scenario, the collector collects many different statistics. All those statistics are identified by 1 collection. If the same test happens again, a different collection record is created.

However, we currently completely ignore this table (it is never used in any read query - only inserts).

It seems that if we run many different test iterations a different collection is created for each iteration.

Questions

Is this useful? Do we care about test results generated from different tests for the same test case and artifact combination?

Start by reading the new glossary in issue #938 and the related schema discussion in issue #937. Inspect how the collection table is currently inserted and confirm that it is not used by read queries. Done means deciding whether repeated test collections are useful and documenting or implementing the agreed meaning and usage.