feat: add llmobs proxy paths to trace agent

[sabrenner]

Adds LLM Observability Eval Metrics and Spans endpoints to the trace agent. They are EVP proxy paths to begin with, and then we forward to our actual backend (proxy path without /evp_proxy/v2).

I verified this by using it as a layer in a small lambda I have for LLM Observability testing - with a debug log I added, I noted

DD_EXTENSION | DEBUG | Got response from https://llmobs-intake.datadoghq.com/api/v2/llmobs backend: 202 Accepted

and was able to see the span in LLM Observability. This was verified for both Python and Node.js.

MLOB-2521

[duncanista]

What are the endpoint limits? What's the max size package? Does the tracer already batch the data accordingly?

[sabrenner]

What are the endpoint limits?

I think the only limitations are payload size, which we have handling for in our SDKs and ingestion paths. Otherwise our intake is JSON based, and all payload encoding is handled in our SDKs. This PR would just add the pass-through endpoints so we can leverage the API key on the agent.

What's the max size package?

I'm guessing you mean the payload size (correct me if I'm wrong!) - we have a 1mb limit on our spans, which we have checks for in the SDKs that will be using this endpoint. We're actually actively reconfiguring this to send spans as batched events (which is actually still just one event from our side, but read as a batch at ingestion).

Does the tracer already batch the data accordingly?

Yeah, even in its current state everything is handled by the tracer, but we are releasing a version of our SDK that submits spans as a batch.

Let me know if this answers your questions @duncanista!