Add Vec::sign_bits

[atlv24]

This is dependent on the fixes in #344 (those problems were found while adding this function)

I need this function for optimized tetrahedron face winding order checks (four dot products being put in a Vec and then grabbing the sign bits to check if its either 0b0000 or 0b1111, thus having all the same winding order and not being degenerate) among other things.

BVec::bitmask almost does this, but gets the LSB for scalar implementations, but I need it on Vec anyways so its irrelevant

[bitshifter]

I'm wondering what the most general purpose method would be for most users here.

For example, there are already methods on VecN for is_finiite -> bool (all elements finite) and is_nan -> bool (any elements are nan). I've tried to match the naming use on f32/f64 there although I guess it's not entirely consistent about whether the result is any or all. There is also is_nan_mask which returns BVecN.

For your usage it seems like a is_sign_negative which returns true if all elements are sign negative would fit your use case? That would probably fit better with the existing naming scheme. Do you need a u32 mask beyond checking if it's all zeros or all ones?

[atlv24]

My use-case needs to distinguish between three cases: all positive, all negative, or mixed signs. I don't need to know what the specific mix is if its mixed. Sample code:

    match sign_winding.sign_bits();
    {
        // All positive
        0 => sign_origin.cmpge(Vec4::splat(-f32::EPSILON)),

        // All negative
        0xf => sign_origin.cmple(Vec4::splat(f32::EPSILON)),

        // Mixed signs, degenerate tetrahedron
        _ => BVec4A::splat(true)
    }

[bitshifter]

Kind of a note to myself, it might be good to add TRUE and FALSE constants to BVecN types for this kind of use.

[atlv24]

Funny, there's a FALSE const for all BVecN types but no TRUE. I can PR this if its convenient

[bitshifter]

The current false is private I think. If you want to make a PR, go for it.