feat: Support Decimal type in approx_distinct

[mkleen]

Which issue does this PR close?

• Relates to Support more types for approx_distinct function #22989 but does not close it. More types are coming.

Rationale for this change

• Support the SQL Decimal type for approx_distinct

• The Arrow types Decimal32, Decimal64, Decimal128 and Decimal256 can be directly supported for NumericHLLAccumulator and HllGroupsAccumulator

What changes are included in this PR?

• Enable NumericHLLAccumulator and HllGroupsAccumulator to support Decimal32, Decimal64, Decimal128 and Decimal256.
• Tests for the non-grouped and grouped path as part of approx_distinct.rst and aggregate.slt for Decimal128 and Decimal256
• Benchmarks

Are these changes tested?

Yes

Are there any user-facing changes?

Yes, approx_distinct supports now Decimal but no breaking changes.

[Jefffrey]

lets add decimal32 & decimal64 as well

[mkleen]

lets add decimal32 & decimal64 as well

AFAIK no SQL Type mapping exists for Decimal32 and Decimal64 see https://datafusion.apache.org/user-guide/sql/data_types.html. Does it make sense to support it then? Following count also just supports Decimal128 and Decimal256.

datafusion/datafusion/functions-aggregate/src/count.rs

Lines 195 to 200 in ff677c4

	DataType::Decimal128(_, _) => Box::new(PrimitiveDistinctCountAccumulator::<
	Decimal128Type,
	>::new(data_type)),
	DataType::Decimal256(_, _) => Box::new(PrimitiveDistinctCountAccumulator::<
	Decimal256Type,
	>::new(data_type)),

[Jefffrey]

i dont think we should base support only on sql type mapping; for example this could be called via dataframe api, or if sql has arrow_cast, etc.

count does support it via the generic catchall, but i think its just an oversight that decimal32/decimal64 dont have specialized paths since they were types added later (compared to decimal128/decimal256)

datafusion/datafusion/functions-aggregate/src/count.rs

Lines 262 to 266 in ff677c4

	// Use the generic accumulator based on `ScalarValue` for all other types
	_ => Box::new(DistinctCountAccumulator {
	values: HashSet::default(),
	state_data_type: data_type.clone(),
	}),

[mkleen]

i dont think we should base support only on sql type mapping; for example this could be called via dataframe api, or if sql has arrow_cast, etc.

count does support it via the generic catchall, but i think its just an oversight that decimal32/decimal64 dont have specialized paths since they were types added later (compared to decimal128/decimal256)

datafusion/datafusion/functions-aggregate/src/count.rs

Lines 262 to 266 in ff677c4

	// Use the generic accumulator based on `ScalarValue` for all other types
	_ => Box::new(DistinctCountAccumulator {
	values: HashSet::default(),
	state_data_type: data_type.clone(),
	}),

ok, sounds good.

[mkleen]

@Jefffrey Thank you for the review!

[Jefffrey]

thanks @mkleen