[SPARK-58948][SQL] Deduplicate nulls-buffer creation in columnar decompressors

[uros-b]

What changes were proposed in this pull request?

Each columnar decompressor repeats the same two-line idiom to build the "nulls buffer" it reads null-tracking state from:

val nullsBuffer = buffer.duplicate().order(ByteOrder.nativeOrder())
nullsBuffer.rewind()

This adds a createNullsBuffer(buffer: ByteBuffer) helper to the CompressionScheme companion object (alongside the existing columnHeaderSize helper) and routes the six call sites through it (PassThrough, RunLengthEncoding, DictionaryEncoding, BooleanBitSet, IntDelta, LongDelta). The subsequent nullCount / nextNullIndex reads stay inline, since the nextNullIndex sentinel legitimately differs (-1, and capacity for PassThrough).

Why are the changes needed?

The idiom was duplicated across all six decoders. Naming it once removes the duplication and gives a single home for the "duplicate, set native order, rewind" convention. It also normalizes the LongDelta site, which called rewind without parentheses.

Does this PR introduce any user-facing change?

No.

How was this patch tested?

Existing columnar compression tests. createNullsBuffer returns exactly the previously-inlined expression, so decompression behavior is unchanged.

Was this patch authored or co-authored using generative AI tooling?

Generated-by: Claude Code (Opus 4.8)

[uros-b]

Thank you for review @dongjoon-hyun

[dongjoon-hyun]

+1, LGTM