Handle 16-bit architectures

[mbrubeck]

Vec32 assumes that casting from u32 to usize is always lossless, which is not true on 8-bit or 16-bit platforms and could cause undefined behavior there.

[setharnold]

How about 'index' larger than 2^31 on 32 bit platforms?

[mbrubeck]

How about 'index' larger than 2^31 on 32 bit platforms?

That would also be bad. Looking closely I don't think it's possible to trigger, because all allocation is done through std::vec::Vec, which panics if it tries to allocate more than isize::MAX bytes. Since cap can never exceed isize::MAX, and len can never exceed cap, the check that index < len is sufficient.