Specialize heap small and large chunk implementations (#4366)

Prior to this commit, the heap `chunk_t` was generic for both small
and large chunks taking up more space than required for either one.

This commit specializes the small and large chunk structures by
using pointer tagging on the `chunk->m` field to keep track of
the chunk type and large chunk slot/shallow/finaliser and the small
chunk sizeclass. This should in theory improve performance as now
the `large_chunk_t` is 32 bytes and `small_chunk_t` is 36 bytes on
64 bit platforms (while `chunk_t` was previously 44 bytes.

There are no logic changes other than whatever was required to
support the specialization.