WIP: Propagate trace context from Cowboy to Plug

[GregMefford]

This is still WIP, but I got the basics working for using @binaryseed's cowboy_telemetry that was recently added to PlugCowboy. I'm hopeful that this will allow us to much more reliably capture traces for all requests, even if something crashes or exits in the Plug request process.

I shoved it in this branch so that I could quickly try to get it working without having to juggle all the Spandex deps, but I think ultimately this belongs in a re-usable library instead of copy/pasted into everyone's apps.

[zachdaniel]

This looks very cool :)

[zachdaniel]

Its interesting, I thought someone ended up implementing some code in spandex_ecto that looked up the trace context from the caller_pid, but I can't find it. But in that case, it should mean that some spans for ecto's parallel preload would just be lost, right? I was thinking the telemetry handler managed ets table could be good solution for the ecto issue as well, but maybe thats a non-issue now 🤷‍♂️ .