Add the vectorization support for masked load when there is out-of-bound exception like tail loop with predicate feature.

Here is the current java implementation (FIXME: optimize part) which is not intrinsified:

    @ForceInline
    public static
    ByteVector fromArray(VectorSpecies<Byte> species,
                                   byte[] a, int offset,
                                   VectorMask<Byte> m) {
        ByteSpecies vsp = (ByteSpecies) species;
        if (offset >= 0 && offset <= (a.length - species.length())) {
            return vsp.dummyVector().fromArray0(a, offset, m);
        }

        // FIXME: optimize
        checkMaskFromIndexSize(offset, vsp, m, 1, a.length);
        return vsp.vOp(m, i -> a[offset + i]);
    }