feat: go-libp2p v0.16.0

[lidel]

This PR aims to fix the interop tests in ipfs/kubo#8522 and close #396

• switch to go-ipfs from WIP: update go-libp2p to v0.16.0 kubo#8522
• refactortest/utils/circuit.js to use new go-ipfs-config keys
• disable JS-GO circuit combinations that are not possible due to missing circuit v2 support in JS
• make node bin/ipfs-interop.js -- -t node --grep "circuit" pass

[lidel]

Refactored GO side of things to leverage circuit v2 and ensure correct relay is used by setting StaticRelays (both added in ipfs/kubo#8522).

Unfortunately JS side of things (js-ipfs running in nodejs or in browser) does not talk circuit v2 and some combinations fail to pass. Landing this in time for go-ipfs 0.11 is unlikely and I think we need to skip them for now – marked all cases with FIXME keyword so we don't forget + will fill issue in js-ipfs repo after this PR is reviewed.

---

For future reference, combinations which currently do not pass:

• terse notation:
  a-r-b – node a talking to node b over relay r
js is js-ipfs in node
  browser is js in a browser context (a regular webpage, for sake of making this list short)
  go is go-ipfs with circuit v2 (acting as a relay client or providing the service)
  💚 cigar (circuit works, both connect and send. note: if JS is the relay, means everyone involved speaks v1)
  💢 no cigar (failure, either connect or send)

• test/circuit/all.js

  • 💚 go-go-go
  • 💢 js-go-go
  • 💢 go-go-js
  • 💢 js-go-js
  • 💚 go-js-go
  • 💚 js-js-go
  • 💚 go-js-js
  • 💚 js-js-js

• test/circuit/browser.js

  • 💢 browser-go-js
  • 💢 browser-go-go
  • 💚 browser-js-js
  • 💚 browser-js-go
  • 💢 js-go-browser
  • 💢 go-go-browser
  • 💚 js-js-browser
  • 💚 go-js-browser
  • 💢 go-browser-browser (was already skipped due to lack of webrtc in go)
  • 💚 js-browser-browser
  • 💚 browser-browser-go
  • 💚 browser-browser-js

If I read this correctly, JS-IPFS cant use GO-IPFS relays v2 (fails in both node and browser scenarios), and needs to use v1 from js-ipfs until support for v2 is added to JS.

[lidel]

Sidenote:

• Circuits that would fail due to JS missing circuit v2 are now skipped
• CI failure is due to unrelated sharding changes in go-ipfs 0.11
  • When we get fix: use new go-ipfs sharding option #393 to pass, this PR should be rebased on top of that.

[marten-seemann]

There are 2 reasons why a test might fail here:

1. JS doesn't speak circuit v2.
2. If a Go node is connected to another peer via a v2 relay, it will only open streams for certain protocols (most notably, the hole punching protocol). That means that although a connection was successfully established, we might not be able to use it to transfer a file.

We probably should have another set of tests using relayd as a relay, configured as a v1 relay. With that configuration, all tests that we currently have should pass.

[lidel]

I've split circuit tests into v1 and v2 sections. v1 tests connect and send, v2 only tests connect.

Tests of circuit v1 and v2 work, there is an unrelated failure in hamt shards test - most likely go-ipfs from update-libp2p pulled in some changes

@achingbrain mind eyeballing this PR?

Note:

• tests for v1 now actually work (previously, go-ipfs was smart enough to ignore relay and connect directly thanks to some combination of hints from DHT/MDNS/Bootstrappers ✨ 💫 )
• kept old tests (some tests had to be disabled because go-ipfs only support v2)
• in node, added new tests for v1 and v2 that use relayd (not sure if worth it, but i guess they can show us when js/go implementation is broken, while reference one still works)
• we don't have JS orchestration for relayd in browsers, js-ipfs provides v1 in those contexts for now (go-ipfs/js-ipfs have a mature setup based on aegir/playwright/ipfsd-ctl where js/go-ipfs is spawned remotely, no such thing exists for relayd and not sure if it is worth the investment, if we expect v2 to land in js-ipfs soon)

[lidel]

CI was failing due to sharding changes, we continue this in #400 (CI is finally green there)