perf(bigquery): [T1-2] zero-allocation column name lookup in FieldList.getIndex

[lqiu96]

Summary & Bottleneck

When callers access row values by column name via FieldValueList.get(String name), every cell lookup delegates to FieldList.getIndex(String name). Previously, FieldList backed nameIndex with a custom CaseInsensitiveMap subclass of HashMap that executed key.toString().toLowerCase() on every get(Object key) call.

• Whenever a column name contained any uppercase character (such as CamelCase or UPPER_SNAKE_CASE), String.toLowerCase() allocated a new String and backing byte array on every cell read across every row ($O(\text{rows} \times \text{cols})$ allocations).
• Even for all-lowercase column names, String.toLowerCase() scanned every character of the column name on every cell access and consulted Locale.getDefault().
• In addition, new CaseInsensitiveMap() initialized the underlying HashMap with default capacity (16), causing repeated table resizing on wide schemas.

Changes

• In java-bigquery/google-cloud-bigquery/src/main/java/com/google/cloud/bigquery/FieldList.java:
  • Pre-size nameIndex using Maps.newHashMapWithExpectedSize(size * 2) and populate both the lowercase column name (name.toLowerCase(Locale.ROOT)) and the exact-case column name at FieldList construction time.
  • Preserve identical resolution semantics when case-insensitive collisions exist.
  • In getIndex(String name), probe nameIndex.get(name) first so exact-case and lowercase column names resolve in a single hash probe with zero character scanning and zero allocations, falling back to nameIndex.get(name.toLowerCase(Locale.ROOT)) only on a miss.
• In java-bigquery/google-cloud-bigquery/src/test/java/com/google/cloud/bigquery/FieldListTest.java:
  • Add unit test coverage in testGetIndexExactLowerAndMixedCase verifying exact-case, lowercase, mixed-case, and missing column name lookups.

Expected Impact

• Eliminates $O(\text{rows} \times \text{cols})$ String and byte[] allocations when reading non-lowercase column names via FieldValueList.get(String).
• Eliminates per-cell character scanning and Locale.getDefault() lookups on exact-case and lowercase column name hits.
• Eliminates intermediate HashMap table resizing during FieldList construction for wide schemas.

Verification

• Ran unit tests in FieldListTest, SchemaTest, and FieldValueListTest:
  mvn test -pl java-bigquery/google-cloud-bigquery -Dtest=FieldListTest,SchemaTest,FieldValueListTest