fix(arrow-select): preserve nullif nulls for REE and Union

[yongster]

nullif creates nulls by adding a top-level validity bitmap. This works for
most arrays, but RunEndEncoded and Union arrays derive logical nullability from
their child arrays instead.

As a result, nullif previously returned non-null values at positions where
the condition was true for these array types.

Closes #10990

• Delegate nullif for RunEndEncoded and Union arrays to take, using null
  indices for the positions selected by the condition.
• Return an error when a RunEndEncoded values field is non-nullable and cannot
  represent the requested nulls.
• Return an error when a Union has no nullable child that can represent null
  indices.
• Prefer a nullable Union child when take needs to represent a null index.
• Add coverage for nullable and non-nullable REE, plus dense and sparse Union
  cases.

[Jefffrey]

Suggested change

	if !values.is_nullable() && right.iter().any(|value| value == Some(true)) {
	if !values.is_nullable() && right.has_true() {

• https://docs.rs/arrow/latest/arrow/array/struct.BooleanArray.html#method.has_true
• https://docs.rs/arrow/latest/arrow/buffer/struct.BooleanBuffer.html#method.has_true

[yongster]

Done in 9ec63d5. The should-null bitmap is computed once (right.values() & validity), then this check uses BooleanBuffer::has_true() on that mask.

[Jefffrey]

perhaps we can reuse the computation we do here:

arrow-rs/arrow-select/src/nullif.rs

Lines 58 to 71 in b372d1f

	// left=0 (null) right=null output bitmap=null
	// left=0 right=1 output bitmap=null
	// left=1 (set) right=null output bitmap=set (passthrough)
	// left=1 right=1 & comp=true output bitmap=null
	// left=1 right=1 & comp=false output bitmap=set
	//
	// Thus: result = left null bitmap & (!right_values | !right_bitmap)
	// OR left null bitmap & !(right_values & right_bitmap)
	// Compute right_values & right_bitmap
	let right = match right.nulls() {
	Some(nulls) => right.values() & nulls.inner(),
	None => right.values().clone(),
	};

instead of needing to check validity in this iterator

[yongster]

Done. should_null is now computed once and reused by both the generic nullif path and nullif_take, so we no longer inspect is_valid/value per index.

[Jefffrey]

• same fixes as fix(arrow-select): preserve nullability for REE and Union take #10994? should we wait for it to land first?

[yongster]

Yes — the take.rs changes here are a subset of #10994. I'll wait for that to land, then rebase this PR so it only contains the nullif.rs changes.

[yongster]

@Jefffrey on the relationship with #10994:

This PR's nullif path for RunEndEncoded and Union delegates to take with null indices. That only produces correct logical nulls if take itself:

• prefers a nullable Union child for a null index
• errors when a REE values field is non-nullable
• errors when a Sparse Union has non-nullable fields

Those take fixes live in #10994. The small take.rs diff here is a subset of that PR, included so this branch still works against current main.

I'd like to wait for #10994 to land, then rebase this PR so it only contains the nullif.rs changes.