providers: don't add every connected node as a provider

[Stebalien]

We now do exactly what the comment is warning about: track peers providing keys.

NOTE: This should be tested against the live network.

[Stebalien]

I believe (untested) the tests are failing because we're never sending out a want request due to a lack of peers. The failing test is TestSessionFindMorePeers.

[hannahhoward]

@Stebalien this looks LGTM to me. But I have no idea the affect it will have in practice. I wonder if we should wait at least until the non-sessions GetBlocks & providerqueryworker is removed (see ProviderQueryManager PR) -- in the case of sessions we only call FindProvidersAsync when we don't have any peers already that are returning the given block, so I agree that it wouldn't make sense to send these, though I don't believe it will do any harm.

TestSessionFindMorePeers does not actually use any kind of a real network (the mock one is right in the test) so I don't think we are affected by this however.

[hannahhoward]

I need to do a test reliability pass -- see #61 -- I didn't realize it was possible to run tests locally multiple times in a row.

[Stebalien]

SGTM. We can sit on this a bit.

[hannahhoward]

@Stebalien based on our experience with Gateway tracing (that this results in a huge number of spurious attempts to connect to peers), I would like to go ahead and merge this after a bit more live testing.

[Stebalien]

SGTM.

Note: I'm not sure if this was causing us to actually connect to many new peers as it returns peers we're already connected to. But, IMO, we should still go with this. Otherwise, sessions don't quite work correctly.