Add BigQuery backend connection testing

[ruslandoga]

Implements O11Y-1947 with a REST-based BigQuery connection test that deterministically selects an attached source—preferring a direct attachment and falling back to a rule source—and inserts one labeled probe event into its existing table.

The probe requires a same-owner source association, performs no dataset or table DDL, and returns source_required before making a BigQuery request when none exists.

REST responses are normalized to atom-only sanitized errors, and the LiveView explains how to satisfy the source prerequisite.

It intentionally does not wait for Storage Write gRPC support.

Tests: the BigQuery adaptor file passes with 35 tests and one property, the backend component file passes with 5 tests, and Dialyzer, formatting, diff checks, and targeted strict Credo pass.

[depthfirst-app]

🟡 Severity: MEDIUM

project_id and dataset_id are user-controlled backend fields, but this call uses Logflare’s shared ingest credential to create _logflare_connection_test in that GCP dataset, even without an attached source. A tenant can target another customer’s project where the shared account is authorized, causing cross-tenant table creation and probe writes.
Helpful? Add 👍 / 👎

💡 Fix Suggestion

Suggestion: Add a project/dataset ownership check at the beginning of test_connection/1 (around line 189) to verify that the user-controlled project_id and dataset_id actually belong to the requesting tenant before invoking any GCP API calls with Logflare's shared ingest credentials.

Specifically:

1. In test_connection/1, extract user_id from the %Backend{} struct pattern match.
2. Fetch the user via Users.Cache.get(user_id) and add a with clause guard that compares project_id (from backend.config) against the user's bigquery_project_id (from the User struct) and also checks dataset_id against user.bigquery_dataset_id. If the values do not match, return {:error, :invalid_config} (or a dedicated :unauthorized error atom) before any GCP call is made.
3. For tenants using Logflare's own shared project (where user.bigquery_project_id is nil), also guard that the provided project_id equals GCPConfig.default_project_id() and the dataset_id matches the expected user-scoped dataset.

This ensures that the shared ingest credential is never used to probe a GCP project/dataset that the requesting tenant does not own.