fix(bigquery-jdbc): normalize timestamp string representation and improve temporal coercions - #14037

b/544843125

This PR standardizes TIMESTAMP string representation across both Arrow and JSON result sets and introduces a unified mathematical epoch parser for temporal coercions.

Problem Statement

1. Engine Inconsistency in getString(): Previously, getString() on the JSON REST path returned raw numeric epoch decimal strings (e.g. "1408452095.22" or scientific notation "1.6905474E9"), diverging from the Arrow engine which returns formatted timestamp strings.
2. Timezone Discrepancies: Legacy timestamp string formatting was sensitive to JVM-default timezones rather than adhering to UTC.
3. Lossy Nanosecond Precision on JSON Path: FieldValueToTimestamp coerced timestamps through getTimestampValue(), truncating precision to microseconds and causing a single-second offset for pre-1970 negative epoch timestamps due to un-floored division.

Changes Made

• TIMESTAMP String Normalization:

  • Updated BigQueryJsonResultSet.getString() to consistently format TIMESTAMP columns into standard UTC timestamp strings ("yyyy-MM-dd HH:mm:ss.ffffff"), matching the Arrow engine while safely delegating RECORD and REPEATED fields.

• Unified Mathematical Temporal Parser:

  • Introduced BigQueryTemporalUtility.parseEpochDecimalToInstant() using BigDecimal mathematical flooring (RoundingMode.DOWN). This losslessly handles standard epoch decimals, scientific notation (e.g. "1.6905474E9"), whole integer epochs, and pre-1970 negative epoch decimals without sign-inversion or off-by-one errors.
  • Added formatTimestampStringFromMicroseconds() using primitive integer arithmetic (Math.floorDiv/Math.floorMod) for high-throughput streaming on Arrow paths.

• Improved Temporal Coercion:

  • Updated FieldValueToTimestamp in BigQueryTypeCoercionUtility to delegate directly to parseEpochDecimalToInstant(), preserving up to nanosecond precision for getTimestamp() and getObject().

• Resource Management:

  • Added try-with-resources in ArrowUtilities.serializeVectorSchemaRoot() to properly close ArrowRecordBatch and prevent buffer leaks in tests.

Testing

• BigQueryTemporalUtilityTest: Added tests for standard epoch decimals, scientific notation, pre-1970 negative timestamps ("-1.5", "-0.123456"), and deterministic truncation (RoundingMode.DOWN).
• FieldValueTypeBigQueryCoercionUtilityTest: Added tests for nanosecond precision preservation, scientific notation, pre-1970 negative epochs, and sub-second truncation.
• BigQueryJsonResultSetTest: Verified that getString() returns standardized UTC formatted timestamp strings for scalar timestamps and handles RECORD/REPEATED fields.
• BigQueryJsonArrayOfPrimitivesTest: Updated test fixture to exact 6-digit timestamp.