buck2_client: download traces via buck2.log_url config key

[thoughtpolice]

When doing a command like buck2 log replay --trace-id ... it helps to be able to download previously created logs from users, automated CI, etc. This functionality exists within Meta, as logs are available for download through the "Manifest" system, but it isn't usable in OSS, despite being useful for diagnostics and getting help.

The only missing thing to really get things working is a download mechanism for log files, and a way to know where they should come from.

With this patch, if a user configures the buck2.log_url key, which is expected to look something like:

[buck2]
log_url = https://example.com

Then, upon executing a log command with a --trace-id flag, this server will be queried with a:

GET /v1/get/{uuid}

request, which is expected to return the raw zst-encoded protobuf file. So, buck2 will do that and download the trace, and then use it like normal.

The request is done with curl, though in principle it could be done within Buck itself.

This shares as much code as possible with the existing infrastructure and tries to only insert a small key set of #[cfg(fbcode_build)] directives. Notably, the path Meta uses for their "Manifold" client is still fully available in the OSS version; fbcode_build is only used to gate what the default choice is.

How the server gets access to these files and how they are uploaded is another question. But for now, this can be done several ways outside of buck2 core, so this is good enough.

GitHub Issue: #441

[thoughtpolice]

This can't do things like pass authentication credentials just yet, but I'd like to know if this is the right shape. As it is, it's still incredibly useful, so I'd like to know if it looks like roughly the right thing before doing much more.

[thoughtpolice]

I'd like to make some small API changes to the GET URI so I'm holding this. But I'd still really like something like this to go in :)