Support dynamic tracing per Tenant/Timeline/...

[shanyp]

Original comment from @kelvich:

It would also be super useful to have:

per-module tags on logs and the ability to enforce per-module log level in a runtime, e.g. through http api. ability to enforce all logs for a specific tenant in runtime depending on complexity that could be done now or later

### Tasks
- [ ] https://github.com/neondatabase/neon/pull/3374

[SomeoneToIgnore]

Status:

• there's an external PR into tracing library to add bits of the required functionality
• there's neondatabase/ a fork of tracing and a PR https://github.com/neondatabase/tracing/pull/1 that mimics the upstream PR
• the PR's branch from the form is used in https://github.com/neondatabase/neon/pull/3374 and has Python IT tests that show that the funcitonality works the way intended: we can dynamically make any log line, labelled with tenant/timeline id to emit or stop emitting logs

The PR misses one crucial thing that I'd love to do somehow: a benchmark comparison against main to see if the performance could've degraded somehow. That could be due to the nature of the chage in the PR into tracing: to make tracing logger to react to filter changes dynamically, it has to make more checks after the patch everywhere where spans are used.

For such a global change, it would be good to have a benchmark with multiple tenant and timelines involved, no extra log filtering features enabled, comparing two versions of tracing. I am not aware of any way to interpret current benchmark results, neither of the way to display them on the charts.