perf(compute): batch contiguous filter take indices

[fallintoplace]

Summary

• Add a range writer for uint16 and uint32 take indices.
• Use it when a filter block or bitmap run is fully selected.
• Keep the scalar path for mixed blocks.
• Add benchmarks for dense, clustered, alternating, short, and nullable filters.

Benchmark

On an Apple M1 Pro, the 1M-row all-selected case improved from about 3.0 ms to 0.5 ms locally. Fragmented filters stayed close to the baseline.

Tests

• go test ./arrow/compute/internal/kernels ./arrow/compute
• go test -race ./arrow/compute/internal/kernels ./arrow/compute
• Full package test run passed with the local Parquet test data checkout, excluding the CSV example that needs a separate Arrow CSV fixture.