Removed ALIGNMENT invariant from [Mutable]Buffer

[jorgecarleitao]

This issue tracks a backward incompatible change whereby buffers are no longer aligned to ALIGNMENT (arch-dependent, 128 in x86_64).

This change allows people using Vec directly with this crate, but removes some of potential for optimizations when e.g. adding two arrays together with load_aligned.

The reason for this change is that the aligned load use-case is narrower than the use-case of being able to use Vec on the crate without an extra memcopy.

To keep the old behavior, use crate feature cache_aligned.