Use BitChunkIterator in Parquet Writer Level Computation

**Is your feature request related to a problem or challenge? Please describe what you are trying to do.**

The level computation logic added in #1588 iterates through null bitmaps, checking each bit individually.

It is possible to iterate the set bits in a buffer more quickly, by instead reading u64 chunks, and counting leading zeros. This filter kernel's `IndexIterator` currently uses this trick, along with the parquet reader with `iter_set_bits_rev`.

**Describe the solution you'd like**

* Extract this bit mask iteration logic so that it can be reused
* Update the parquet writer to use this trick

**Describe alternatives you've considered**

We could not do this

**Additional Context**

Similar to https://github.com/apache/arrow-rs/issues/1600

Start by comparing the filter kernel's IndexIterator and the Parquet reader's iter_set_bits_rev with the Parquet writer's level computation logic. Extract the shared set-bit iteration behavior so it can be reused, then update the writer to use u64 chunks and leading-zero counting. Done means the writer uses the reusable iterator and its existing behavior remains correct.

Written by the indexing model from the issue text.