[lqiu96]

Summary & Bottleneck

• In BigQueryImpl.queryRpc, when Page 1 completes with a non-null pageToken, new QueryPageFetcher(...) executes a synchronous blocking this.job = getJob(jobId) HTTP RPC inside its constructor before returning Page 1 to the caller. If the caller only inspects Page 1, the jobs.get RPC is wasted, and even for multi-page consumers it adds a full HTTP round-trip to time-to-first-row. QueryPageFetcher also extends java.lang.Thread without ever starting the thread, registering an unstarted thread in the active ThreadGroup on every multi-page query.
• In ConnectionImpl.processQueryResponseResults, getDestinationTable(jobId) executes a synchronous blocking jobs.get HTTP RPC even when results.getPageToken() == null (single-page fast-path queries) and submits an unnecessary background pagination task.
• In ConnectionImpl.getSubsequentQueryResultsWithJob and tableDataList, Schema.fromPb(firstPage.getSchema()) is parsed twice and getQueryJobRpc(jobId) (jobs.get HTTP RPC) is invoked up to three times for the same JobId.

Changes

• Remove extends Thread from QueryPageFetcher and defer getJob(jobId) and destinationTable lookup lazily into QueryPageFetcher.getNextPage() in java-bigquery/google-cloud-bigquery/src/main/java/com/google/cloud/bigquery/BigQueryImpl.java.
• Guard getDestinationTable(jobId) and runNextPageTaskAsync behind pageToken != null in processQueryResponseResults and tableDataList in java-bigquery/google-cloud-bigquery/src/main/java/com/google/cloud/bigquery/ConnectionImpl.java.
• Deduplicate Schema.fromPb in getSubsequentQueryResultsWithJob and cache the last resolved Job per JobId in getQueryJobRpc in java-bigquery/google-cloud-bigquery/src/main/java/com/google/cloud/bigquery/ConnectionImpl.java.
• Add a unit test in java-bigquery/google-cloud-bigquery/src/test/java/com/google/cloud/bigquery/ConnectionImplTest.java verifying that single-page processQueryResponseResults skips getDestinationTable and getQueryJobSkipExceptionTranslation.

Expected Impact

• Saves 1 full synchronous HTTP round-trip (~50 to 150 ms) on every single-page Connection fast-path query and before Page 1 return on every multi-page BigQuery.query(...) call.
• Saves 1 to 2 redundant jobs.get HTTP round-trips and 1 duplicate Schema.fromPb deserialization on Connection fallback and Read API queries.
• Eliminates unstarted java.lang.Thread allocations and ThreadGroup.addUnstarted() overhead in QueryPageFetcher.

Verification

• Ran ConnectionImplTest and BigQueryImplTest via Maven (174 tests run, 0 failures).