Add nightly support for more SIMD implementations

[mooman219]

There are unstable APIs for SIMD on platforms like Web, ARM, and POWER. Ideally add a way to utilize this APIs if the user is on nightly.

[john01dav]

https://blog.rust-lang.org/inside-rust/2020/09/29/Portable-SIMD-PG.html

It might be better to use this than to support lots of platforms manually. That way you get support for more platforms (presumably) with less work.

[mooman219]

@john01dav There's some spicy bits that will need hand written intrinsics. See:

fontdue/src/platform/float/get_bitmap.rs

Lines 38 to 61 in 2bfda9d

	// x = Read 4 floats from self.a
	let mut x = _mm_loadu_ps(a.get_unchecked(i));
	// x += (0.0, x[0], x[1], x[2])
	x = _mm_add_ps(x, _mm_castsi128_ps(_mm_slli_si128(_mm_castps_si128(x), 4)));
	// x += (0.0, 0.0, x[0], x[1])
	x = _mm_add_ps(x, _mm_castsi128_ps(_mm_slli_si128(_mm_castps_si128(x), 8)));
	// x += offset
	x = _mm_add_ps(x, offset);
	// y = x * 255.9
	let y = _mm_mul_ps(x, _mm_set1_ps(255.9));
	// y = abs(y)
	let y = _mm_andnot_ps(_mm_castsi128_ps(nzero), y);
	// y = Convert y to i32s and truncate
	let mut y = _mm_cvttps_epi32(y);
	// y = Take the first byte of each of the 4 values in y and pack them into
	// the first 4 bytes of y.
	y = _mm_packus_epi16(_mm_packs_epi32(y, nzero), nzero);
	// Store the first 4 u8s from y in output.
	let pointer: &mut i32 = core::mem::transmute(output.get_unchecked_mut(i));
	*pointer = core::mem::transmute::<__m128i, [i32; 4]>(y)[0];
	// offset = (x[3], x[3], x[3], x[3])
	offset = _mm_set1_ps(core::mem::transmute::<__m128, [f32; 4]>(x)[3]);