@repo/gateway package and AiCallLog schema so every future model call can be logged with:traceId
runId
AI call kind
status
provider
model
modelVersion
promptVersion
schemaVersion
input/output JSON
parsed output
error type
latency
tokens
estimated cost
metadata