Print sources for errors in RetryError

Fixes #958 (closed)

I think this has come out very nicely.

Things that may be controversial and ought to be considered:

• Is retry-error the right place for fmt_error_with_sources? Arguably not, because it's really nothing to do with retries. But we don't have a better place to put it; retry-error ought not to depend on tor-error, I think. If we come up with a more-fundamental place, then retry-error could re-export it with #[doc(hidden)] etc., so I think this is an OK thing to do to the API now.

• Whether RetryError ought to print sources at all. Discussion: #958 (comment 2922760) et seq. As I write there, I'm fairly convinced that it should, at least pending some more substantial overhaul.

With this MR, the example error in #958 (closed) looks like the desired-by-me output as I show here: #958 (comment 2922761)