BigQuery exporter for usage and audit logs

Description

There is no BigQuery sink: no BigQuery client and no use of the Storage Write API anywhere in the tree. The GCP-adjacent paths today are the object-store NDJSON usage export and the scheduled CSV audit export, so reaching BigQuery means landing NDJSON in GCS and defining external tables over it.

Requested: a first-class BigQuery exporter over the Storage Write API —

• dataset and table configuration, service-account or Workload Identity auth;
• batched append with retry and backpressure, behind the same async-batched exporter pipeline as the other sinks;
• a documented, stable schema for usage records (model, caller, timestamps, token breakdown, cost, latency, status) and for audit records.

Why

For organisations on GCP, BigQuery is where token-spend chargeback, per-team cost attribution and audit queries already live — every other data source they have lands there directly. Routing through object storage plus external tables adds a hop, hands the operator a pipeline to own, and lets schema drift break things quietly.

This is a sibling of the other named sinks (#57, #62) and would slot into the pluggable-sink work tracked in api7/AISIX-Cloud#692.

Priority

High.

Prior art

No peer clearly ships a first-class BigQuery sink; the two closest are unconfirmed for LLM traffic.

_{Surveyed 2026-07-15; every claim rests on a fetched docs/source page.}