Add integration tests for collectors

[shonfeder]

This adds

• A simple signal gathering server that can be run to collect signals sent by an application instrumented with a collector.
• Changes to the lwt and ocurl emitters to bring them into parity and support testing
• An integration test harness for our otel collectors, using this server.
• Integration tests for the cotthp-lwt collector.

Initial, obviated discussion from when PR was opened as a draft

This relies on a pair of hacks for getting deterministic output of collected signals, which are then used for expect-style testing.

These hacks are sufficient (locally, at least) for the cohttp-lwt emitter, but still yield non-deterministic output for the ocurl emitter. I have in mind two routes for proper testing here:

1. Abandon the expect-based approach. While it seems convenient at first, it is really not suitable for this domain, because
   • (a) the outputs contain non-deterministic data and
   • (b) outside of the trivial case, there is way too much output for the diffs to be very useful. Instead, we can feed the collected signals into an alcotest suite, and just check properties of the collected data.
2. Keep to the expect-based approach, despite the reservations of (1), but project the signal data into a flattened form, and apply an ordering to get determinstic output.

I'm opening this as a draft for now, in case anyone is motivated to share opinions on the approach so far, or suggest other ones.

[c-cube]

I think 1. could make sense if we have some testing utils to "traverse" a collection of signals. We don't need to encode them either, we can just use the Opentelemetry OCaml values. It's certainly the most robust way to ignore the nondeterministic parts.

[shonfeder]

Excellent, this makes the most sense to me too. Thanks!

I am porting the testing server to lwt, to address that part of the issue, and I think I have a clear sense of how to approach the other issue. Thanks!

[shonfeder]

Thanks @c-cube! I've made the changes using the approach we agreed was preferable. In #98 you can see how we will use the exact same conformance tests for the WIP Eio collector.

[c-cube]

I think we could use containers or astring as a test dep, for this kind of issue we see in CI. Or we can bump the minimal OCaml version to 4.13 at least.

[shonfeder]

Yeah, that is nicer than copy pasta of these functions into the tests code :D .

I've opted for Containers, and done a little cleanup pass, and CI is green!

[c-cube]

Excellent, thank you!

[shonfeder]

Thank you for the review!