perf(bigquery): [T2-3] support typed FieldValue backing to avoid binary-to-String-to-binary round-trips in ArrowDeserializer

[lqiu96]

Summary & Bottleneck

In java-bigquery/google-cloud-bigquery/src/main/java/com/google/cloud/bigquery/ArrowDeserializer.java (arrowVectorToFieldValue), every binary primitive extracted from an Apache Arrow vector is eagerly converted into a String and wrapped in FieldValue.of(Attribute.PRIMITIVE, stringVal):

1. Redundant binary-to-String-to-binary round-trips on $O(\text{rows} \times \text{cols})$ cells:
   • IntVector and BigIntVector values are formatted via Integer.toString or Long.toString, only for FieldValue.getLongValue() to immediately re-parse the string via Long.parseLong.
   • Float8Vector values are formatted via Double.toString, only for FieldValue.getDoubleValue() to re-parse the string via Double.parseDouble.
   • TimeStampVector epoch microsecond values are formatted via formatTimestampMicros(long) into "<seconds>.<microseconds>", only for FieldValue.getTimestampValue() to construct and scale a BigDecimal to recover the original long microseconds.
   • DecimalVector (BigDecimal) values are formatted via toPlainString(), only for FieldValue.getNumericValue() to construct new BigDecimal(String).
   • VarBinaryVector (byte[]) values are Base64-encoded via BaseEncoding.base64().encode(byte[]), only for FieldValue.getBytesValue() to Base64-decode them back into byte[].

Changes

• java-bigquery/google-cloud-bigquery/src/main/java/com/google/cloud/bigquery/FieldValue.java:
  • Add package-private typed factory methods (ofLong, ofDouble, ofTimestampMicros, ofNumeric, ofBytes) that retain the typed value in rawValue.
  • Return the typed value directly in getLongValue(), getDoubleValue(), getTimestampValue(), getTimestampInstant(), getNumericValue(), and getBytesValue() without allocating or parsing a String.
  • Lazily materialize and cache the canonical String representation in getValue() and getStringValue() (and use getValue() in equals(), hashCode(), and toString()) so string access and equality checks remain 100% compatible with string-backed FieldValue instances.
• java-bigquery/google-cloud-bigquery/src/main/java/com/google/cloud/bigquery/ArrowDeserializer.java:
  • Construct FieldValue instances using ofTimestampMicros, ofLong, ofDouble, ofBytes, and ofNumeric in arrowVectorToFieldValue.

Expected Impact

• Eliminates $O(\text{rows} \times \text{cols})$ intermediate String formatting, Base64 encoding/decoding, and Long/Double/BigDecimal re-parsing when reading typed columns from Arrow streams via getLongValue(), getDoubleValue(), getTimestampValue(), getNumericValue(), and getBytesValue().
• Preserves full backwards compatibility when callers invoke getStringValue(), getValue(), equals(), or hashCode().

Verification

• Added unit tests in java-bigquery/google-cloud-bigquery/src/test/java/com/google/cloud/bigquery/FieldValueTest.java and java-bigquery/google-cloud-bigquery/src/test/java/com/google/cloud/bigquery/ArrowDeserializerTest.java.
• Ran mvn test -pl java-bigquery/google-cloud-bigquery -Dtest=FieldValueTest,FieldValueListTest,ArrowDeserializerTest,ArrowQueryPageFetcherTest,ConnectionImplTest (59 tests run, 0 failures).