percentile_cont interpolation causes `NaN` for f16 input

[Jefffrey]

Describe the bug

(Relies on #18944 being merged so it can accept f16 input)

This interpolation step here seems to cause f16 to go out of bounds (into inf) so we can't calculate percentiles for f16 as it'll always return NaN:

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

Lines 753 to 758 in 769f367

	let interpolated = lower_value.add_wrapping(
	diff.mul_wrapping(T::Native::usize_as(
	(fraction * INTERPOLATION_PRECISION as f64) as usize,
	))
	.div_wrapping(T::Native::usize_as(INTERPOLATION_PRECISION)),
	);

Need to fix this for f16.

To Reproduce

No response

Expected behavior

No response

Additional context

No response