Capture all telemetry.

[atanikan]

For the inference service, we currently have logging enabled in multiple places, but we don’t actively track most of it.
Right now, the warehouse only ingests database logs for the UI and API. For deeper visibility, it might be useful to centralize and stream additional logs and metrics into Kafka (or another log aggregation pipeline). For example:
a. Stream PostgreSQL and Nginx logs from the VM
b. Stream Globus compute endpoint logs from the Sophia login node
c. Stream vLLM (model) logs from the Sophia login node
d. Stream logs from the OpenWeb UI VM and its PostgreSQL database
e. Collect and forward CPU/Memory/Filesystem usage telemetry from both VMs
f. Perhaps even have GPU telemetry consumed at the same place.
This way, we’d have a more complete picture of system activity without having to enable/capture logs ad hoc.