fix(buffer): keep map_each_in_place output aligned to the mapped type

[jackylee-ch]

What

map_each_in_place transmuted BufferMut<T> into BufferMut<R> but kept T's
alignment. Mapping to a more strictly aligned type (e.g. [u8; 4] → u32) left
the result claiming the weaker alignment, so reading it back as R was UB.

Fix

After mapping, raise the alignment to Alignment::of::<R>() via the existing
aligned helper — metadata-only when the data is already aligned, so the common
same-alignment path is unchanged. This requires R: Copy, which every caller
already satisfies.

The regression test is the one from the issue.

Fixes #9746

AI assistance

Prepared with AI assistance; I have reviewed and verified the change.

[robert3005]

This function is going to get removed... it's used in one place and we have better versions of this logic

[jackylee-ch]

Thanks for flagging. This fixes the under-alignment UB from #9746 (filed with a failing test), so it closes that soundness hole while map_each_in_place is still around. Happy to close this if the removal is imminent or you'd rather fold the fix into that work — just let me know.

[jackylee-ch]

map_each_in_place is slated for removal with better replacements, so I'll close this rather than point-patch a function on its way out. The under-alignment UB stays tracked in #9746 for the replacement to cover. Thanks for the context!