Openinference Instrumentation Proof-Of-Concept

[CalebCourier]

This PR adds OpenInference's GuardrailsInstrumentation on top of the exist FlaskInstrumentation for OTEL. The spans captured by this new instrumentation can be fed to any OTEL-compliant collector including Arize's Phoenix.

.env file with the OTEL settings I was using

OTEL_PYTHON_TRACER_PROVIDER=sdk_tracer_provider
OTEL_SERVICE_NAME=my-service
OTEL_TRACES_EXPORTER=otlp
OTEL_INSTRUMENTATION_HTTP_CAPTURE_HEADERS_SERVER_REQUEST="Accept-Encoding,User-Agent,Referer"
OTEL_INSTRUMENTATION_HTTP_CAPTURE_HEADERS_SERVER_RESPONSE="Last-Modified,Content-Type"
OTEL_METRICS_EXPORTER=none
OTEL_EXPORTER_OTLP_PROTOCOL=grpc
OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4317

Notes on Productionization

1. Since this requires another third party package, we probably want to make this instrumentation optional. That means switching the dependency to a pip extra, performing an optional import in guardrails_api/otel/traces.py, and then only instrumenting when the package is installed. This pip extra would also need to be reflected in the core guardrails-ai package for users utilizing guardrails start.
2. We still need to collaborate with Arize to see how we can/should attach the span attributes that they collect in other instrumentations. See the notion task for more info: https://www.notion.so/guardrailsai/Arize-integration-4fd0096121094cc2896a6200786fb7b0?pvs=4