perf(arrow/array): batch dictionary empty appends

[fallintoplace]

What does this change?

• AppendEmptyValues used to build an n-element values array and dictionary-encode every empty value.
• Resolve the empty value once, reserve the index builder once, and append the same dictionary index for the whole batch.
• Keep the existing null dictionary path unchanged.

Benchmark

Command:

go test ./arrow/array -run '^$' -bench '^BenchmarkDictionaryBuilderAppendEmptyValues$' -benchmem -benchtime=200ms -count=3

Apple M1 Pro, Go 1.26.3. The benchmark reuses the dictionary builder and resets the output length between iterations.

Type	Count	Before ns/op	After ns/op	Before B/op	After B/op
int32	64	1,684	653	1,168	784
int32	4,096	81,910	14,193	42,576	784
int32	65,536	1,305,061	221,358	551,382	784
string	64	2,573	868	1,568	1,056
string	4,096	121,191	14,577	42,912	1,056
string	65,536	1,916,937	230,499	551,714	1,056

Tests

• go test ./arrow/array -count=1
• go test ./arrow/... ./internal/...

[zeroshade]

The batching change preserves behavior in the tested cases and reproduces the expected performance and allocation improvements. Array, repeated dictionary, race, vet, and LSP checks pass. One lint failure must be fixed before merge.

[zeroshade]

The repository’s lint hook rewrites this to b.idxBuilder.Reserve(n), causing the required Lint check to fail because the hook modifies the file. Please use the promoted method and rerun lint.