perf(arrow/array): optimize bulk appends for struct builders

[fallintoplace]

Summary

• reserve the struct and its children once for the full batch
• update the parent validity bitmap and counters once
• dispatch one bulk null or empty append to each child builder
• preserve the existing Go behavior where null structs append null child values
• add mixed nested-struct coverage and width-scaled benchmarks

This uses the same bulk-dispatch orientation as Arrow C++ StructBuilder, while keeping Go child-null semantics unchanged.

Benchmark

65,536 appended structs, representative medians on Apple M1 Pro:

Fields	Operation	Before	After	Change
1	nulls	457 us	245 us	-46%
1	empty	639 us	301 us	-53%
4	nulls	1.40 ms	0.82 ms	-41%
4	empty	1.93 ms	1.07 ms	-44%
16	nulls	5.27 ms	2.86 ms	-46%
16	empty	6.96 ms	3.86 ms	-45%
64	nulls	17.37 ms	11.20 ms	-36%
64	empty	27.31 ms	15.25 ms	-44%

Allocations per build also drop:

• 1 field: 48 to 23
• 4 fields: 132 to 53
• 16 fields: 470 to 175
• 64 fields: 1,846 to about 687

Single-value performance stays about the same.

Tests

• go test -p 1 ./...
• go test -race ./arrow/array -run TestStructBuilderBulkAppendNullsAndEmptyValues -count=1
• go vet ./arrow/array