Revisit caching entry points

[Jason R. Coombs]

While investigating the issue reported in 1941991, I did some work in the `_cache` module, and I noticed that it's primarily there to cache entry points. I don't know the motivation for caching these entries, but I wanted to float the idea that maybe caching the entries is no longer needed.

In the [later releases of importlib_metadata](https://importlib-metadata.readthedocs.io/en/latest/history.html), performance of loading entry points has been dramatically improved, including much faster discovery/loading and an in-memory cache for entry points for a running process.

These performance gains are measured and tracked in the CI runs for importlib_metadata.

If possible, this project should consider removing its own cache in favor of the built-in caching in importlib_metadata (also importlib.metadata in Python 3.10). Doing so should also address #1627906 and would limit the dependencies on the data model (such as construction of EntryPoint objects).

[Stephen Finucane (stephenfinucane)]

This isn't really a bug more so than a feature request, so I'm going to close this. However, we'll probably do this and I've proposed a patch at [1] to get the conversation going.

[1] https://review.opendev.org/c/openstack/stevedore/+/860111