fix(bigquery-analytics): route Storage Write API appends to the dataset's region

[caohy1988]

Summary

Fixes cross-region BigQuery logging failures reported in GoogleCloudPlatform/BigQuery-Agent-Analytics-SDK#262: logging works when the destination dataset is in the US multiregion (e.g. us-central1) but fails with a "session not found" error in single regions such as northamerica-northeast1.

Root cause

The plugin writes rows via the Storage Write API's streaming AppendRows RPC:

responses = await self.write_client.append_rows(requests_iter())

AppendRows is a client-streaming RPC, so the GAPIC client cannot auto-populate the implicit request-routing header (it has no unary request body to read write_stream from). Without that header, every append is routed to the US multiregion by default:

• us-central1 resolves within the US multiregion → the _default stream is found → works.
• northamerica-northeast1 is a single region → the US-routed backend has no knowledge of the stream created there → "session not found" / stream-not-found, the append fails, and no rows are ever written.

This surfaced to users as session_id (and every other column) "failing to propagate" — in reality the rows never landed because the writes were misrouted. The session_id plumbing itself is fine.

Per the Storage Write API docs: "if you write data to any region except the US multiregion, you must include the following header in your requests: x-goog-request-params: write_stream=<stream_name>." The official google.cloud.bigquery_storage_v1.writer.AppendRowsStream helper sets exactly this header (writer.py); the raw async append_rows this plugin calls does not.

Fix

Set the routing header explicitly on the append_rows call so the request reaches the region that owns the write stream:

responses = await self.write_client.append_rows(
    requests_iter(),
    metadata=(
        ("x-goog-request-params", f"write_stream={self.write_stream}"),
    ),
)

US-multiregion behavior is unchanged (the header is correct there too). This is a one-line functional fix that mirrors Google's own client-library helper.

Tests

• Adds test_append_rows_sets_regional_routing_header (deploys the plugin with location="northamerica-northeast1", asserts the routing header is passed to append_rows).
• Full plugin suite green: 229 passed.

Notes / out of scope

• A regional/REP endpoint (bigquerystorage.<location>.rep.googleapis.com) is only required for strict data-residency guarantees; the routing header alone resolves the functional cross-region failure here. Can be a follow-up if data-residency support is desired.
• The constructor's location param is intentionally not passed to the synchronous bigquery.Client — TestDatasetLocationHandling documents that omitting it lets BigQuery infer location from the dataset and avoids silent view-creation failures for non-US datasets. Left as-is.