Default OpenTelemetry sink by ENV variable

[codefromthecrypt]

🚀 Describe the new functionality needed

Let's support the following two variables for the OTLP endpoint when generating configuration, instead of generating a constant.

OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4318
OTEL_EXPORTER_OTLP_PROTOCOL=http/protobuf

The latter we can skip if we only want to support http, which I'm ok with.

💡 Why is this needed? What if we don't build it?

Right now, we can select otel via TELEMETRY_SINKS=otel and control its service name via OTEL_SERVICE_NAME. However, you can't change the OTLP endpoint used without writing your own configuration file, as it is hard-coded to not read a variable.

This implies that first timers will have to edit configuration when using the easiest way to start (docker) because in a container localhost isn't going to the host's localhost.

By supporting ENV defaults, they can change config like this:

OTEL_EXPORTER_OTLP_ENDPOINT=http://host.docker.internal:4318

Other thoughts

A workaround is to use --add-host=localhost:host-gateway which can help folks use ollama and otel when both are running on the host docker process like so:

docker run --rm --name llama-stack --tty -p 5000:5000  \
  --add-host=localhost:host-gateway \
  llamastack/distribution-ollama \
 --env INFERENCE_MODEL=meta-llama/Llama-3.2-3B-Instruct \
 --env TELEMETRY_SINKS=otel