[JasonLi-cn]

Which issue does this PR close?

Closes #.

Rationale for this change

Both ColumnarValue and ScalarValue contain the cast_to function, but they are inconsistent:

datafusion/datafusion/expr-common/src/columnar_value.rs

Lines 196 to 216 in 7b2284c

	ColumnarValue::Scalar(scalar) => {
	let scalar_array =
	if cast_type == &DataType::Timestamp(TimeUnit::Nanosecond, None) {
	if let ScalarValue::Float64(Some(float_ts)) = scalar {
	ScalarValue::Int64(Some(
	(float_ts * 1_000_000_000_f64).trunc() as i64,
	))
	.to_array()?
	} else {
	scalar.to_array()?
	}
	} else {
	scalar.to_array()?
	};
	let cast_array = kernels::cast::cast_with_options(
	&scalar_array,
	cast_type,
	&cast_options,
	)?;
	let cast_scalar = ScalarValue::try_from_array(&cast_array, 0)?;
	Ok(ColumnarValue::Scalar(cast_scalar))

datafusion/datafusion/common/src/scalar/mod.rs

Lines 2822 to 2829 in 7b2284c

	pub fn cast_to(&self, data_type: &DataType) -> Result<Self> {
	let cast_options = CastOptions {
	safe: false,
	format_options: Default::default(),
	};
	let cast_arr = cast_with_options(&self.to_array()?, data_type, &cast_options)?;
	ScalarValue::try_from_array(&cast_arr, 0)
	}

What changes are included in this PR?

unify cast_to of ColumnarValue and ScalarValue.

Are these changes tested?

Are there any user-facing changes?

[alamb]

Thank you @JasonLi-cn -- this makes sense to me. Nice find

[alamb]

🚀