More #[inline] in HalfFloatSliceExt, leading_zeros

[encounter]

This helps code generation for situations with fixed size arrays:

let src: [f16; 4];
let mut dst = [0f32; 4];
src.convert_to_f32_slice(&mut dst);
// ^ This will get optimized to a direct f16x4_to_f32x4

Without the inline modifier, convert_to_f32_slice is always generated as a generic function across arbitrary length slices.