Which issue does this PR close?

• Closes Optimize NULL handling in some datetime functions #21476.

Rationale for this change

to_time, to_local_time, and make_date build NULLs arrays incrementally. It is faster and more idiomatic to build them in bulk via NullBuffer instead.

This doesn't make a huge performance difference in practice, but it shouldn't be slower, and the result is more concise and idiomatic.

Benchmarks (ARM64):

  - make_date_col_col_col_8192: 97.0µs -> 97.3µs, +0.3%
  - make_date_scalar_col_col_8192: 97.3µs -> 97.8µs, +0.5%
  - make_date_scalar_scalar_col_8192: 98.1µs -> 98.7µs, +0.6%
  - make_date_scalar_scalar_scalar: 162.0ns -> 162.1ns, +0.1%
  - to_time_no_nulls_100k: 17.1ms -> 17.7ms, +3.5%
  - to_time_10pct_nulls_100k: 16.0ms -> 16.2ms, +1.3%
  - to_local_time_no_nulls_100k: 8.0ms -> 7.8ms, -2.5%
  - to_local_time_10pct_nulls_100k: 7.6ms -> 7.0ms, -7.9%

What changes are included in this PR?

• Add benchmarks for to_time and to_local_time
• Use 8192 batch size in make_date benchmark, for consistency with default DF batch size
• Implement NULL handling optimization/cleanup

Are these changes tested?

Yes.

Are there any user-facing changes?

No.