Atomics from special heap

[Former user]

We currently caution that once memory has been atomic'd it can no longer be used in any non-atomic op, and therefor cannot be free'd.

Paul suggested we provide a separate memory pool for atomic-able memory.

[Dan Bonachea]

In order to be safe in general, this would need to be a type-specific pool (eg to handle a struct type where some of the fields are accessed atomically and others are not)

Even with type info, it doesn't handle a case like an array of int64_t's where some elements are managed atomically (eg those on a domain border) and others are managed via load/store.

The user has the ability to write a freelist with app-specific knowledge to recycle memory while following the rules. Assuming we end up shipping with the current atomic restrictions, we should probably just encourage them do that.

[Dan Bonachea]

I think this issue is resolved.

Otherwise please re-open.