Fix unaligned memory access with memcpy

[onitake]

As the title says.

Tested on amd64 and sparc64 with gcc 15. All unit tests pass.
The compiled code doesn't actually contain a call to memcpy on either architecture, this is optimized away.

Fixes: #26

[ilevkivskyi]

The procedure is to make any PRs to mypy first, see README, please move your PR there. We can discuss details in new PR (FWIW I think this is a too big change, there are only couple problematic uses of _READ/_WRITE, it is probably better to add a separate pair, say _READ_STRICT/_WRITE_STRICT, for those couple places)

[onitake]

Here is the new PR: python/mypy#20474
I also re-reported the bug on the mypy bug tracker: python/mypy#20473