feat: add observability plugin system

[ajgray-stripe]

Description

See #203 for background; summary of that issue is that we'd like to have a somewhat more abstracted plugin system for observability/tracing so that we can have our own house-built telemetry plugin. I ultimately opted for the relatively lightweight option of just copying Langfuse's API for this PR; I wasn't confident enough in a refactor that was any more different.

In particular, the substantive changes made here are:

• Creates an "observer" plugin system modeled off the existing plugin systems in this repo
• That includes talking with profiles.yaml, pyproject-based entrypoint discovery for plugins, etc.
• Pulling out the langfuse's integration observe_wrapper decorator into a method that calls all registered observers
• Mostly moves current langfuse_wrapper.py code into observers/langfuse.py; I didn't really touch the code in there much. Also makes that plugin a default
  Review commit-by-commit for the logical flow here.

Testing

• I wrote a few unit tests. I checked that both goosewide and exchange tests all passed locally.

I also did some manual testing, with a couple minor hacks to get around not actually being able to test fully end-to-end with Langfuse (the environment I'm using didn't make that easy+fast so I just did this instead).

• With the following temporary patches on top of this branch:
  
  disabling actually authenticating Langfuse

and instead just logging to a file for visibility

, as well as this profile:

we can see that Goose runs correctly, with startup messages showing that it's correctly (trying to) initialize the Langfuse client:

and we correctly log method calls.

• Without the --tracing flag, goose runs without errors as expected: