perf(bigquery): [T2-2] stream Jackson tokens directly into FieldValueList without intermediate TableRow

[lqiu96]

Summary & Bottleneck

Even after PR #4 (perf/bigquery-unwrap-cell-json) unwraps TableCell JSON envelopes in HttpBigQueryRpc, parseTableRows (java-bigquery/google-cloud-bigquery/src/main/java/com/google/cloud/bigquery/spi/v2/HttpBigQueryRpc.java:L2205-L2260) still allocates a new TableRow() (TableRow + GenericData's RegularEnumSet + ArrayMap = 3 objects / 80 bytes per row), a transient ArrayList<Object> per row in parseRowCells, a transient Collections.singletonMap("f", ...) + ArrayList<Object> per RECORD cell, and a transient ArrayList<Object> per REPEATED cell. Immediately afterward, BigQueryImpl.transformTableData (java-bigquery/google-cloud-bigquery/src/main/java/com/google/cloud/bigquery/BigQueryImpl.java:L2220-L2233) and ConnectionImpl.getIterableFieldValueList (java-bigquery/google-cloud-bigquery/src/main/java/com/google/cloud/bigquery/ConnectionImpl.java:L826-L839) iterate over result.getRows(), allocate a second ArrayList<FieldValue> + two Iterators + ImmutableList.copyOf per row in FieldValueList.fromPb, and discard all intermediate TableRow, ArrayList<Object>, and SingletonMap wrappers.

Changes

• Stream Jackson JsonParser tokens in HttpBigQueryRpc.parseTableRows / parseRowCells / parseCellFieldValue (java-bigquery/google-cloud-bigquery/src/main/java/com/google/cloud/bigquery/spi/v2/HttpBigQueryRpc.java) directly into pre-sized ImmutableList<FieldValue> rows and nested FieldValueList values without allocating intermediate TableRow, ArrayList<Object>, or SingletonMap instances.
• Wrap parsed row cell lists in a lightweight per-page StreamedTableRowList carrier (16 bytes per page) and expose @InternalApi HttpBigQueryRpc.extractFieldValueLists(Iterable<TableRow>, Schema, boolean) so BigQueryImpl.transformTableData and ConnectionImpl.getIterableFieldValueList wrap each pre-built ImmutableList<FieldValue> directly into a FieldValueList in zero-copy $O(1)$ per flat row (resolving RECORD sub-field schemas and RANGE element types only when present in the schema), while StreamedTableRowList.get(int) lazily materializes TableRow if accessed by external callers.
• Reuse static NULL_PRIMITIVE and NULL_PRIMITIVE_INT64_TIMESTAMPS flyweights in FieldValue.of (java-bigquery/google-cloud-bigquery/src/main/java/com/google/cloud/bigquery/FieldValue.java) and short-circuit FieldValueList.withSchema(FieldList) (java-bigquery/google-cloud-bigquery/src/main/java/com/google/cloud/bigquery/FieldValueList.java) when the schema is already attached.
• Add unit tests in HttpBigQueryRpcTest (java-bigquery/google-cloud-bigquery/src/test/java/com/google/cloud/bigquery/spi/v2/HttpBigQueryRpcTest.java) verifying HttpBigQueryRpc.extractFieldValueLists across primitive, null, repeated, record, and repeated-record cells.

Expected Impact

• Complexity / Scaling: $O(\text{rows} \times \text{cols})$.
• Eliminates 3 objects (TableRow + RegularEnumSet + ArrayMap = ~80 bytes) per row, 2 intermediate ArrayList + Object[] buffers per row, 2 Iterators per row, intermediate SingletonMap("f", ...) and ArrayList<Object> wrappers on RECORD and REPEATED cells, and all FieldValue allocations on NULL cells.

Verification

• Verified with mvn test -pl java-bigquery/google-cloud-bigquery -Dtest=HttpBigQueryRpcTest,FieldValueTest,FieldValueListTest,BigQueryImplTest,ConnectionImplTest -Dclirr.skip=true -Denforcer.skip=true -Dcheckstyle.skip=true (267 tests passing).