Performance idea for Rust-side bundlers (Drop performance)

[kdy1]

Note: I assume people are using malloc other than snmalloc, which says

snmalloc's design is particular well suited to the following two difficult scenarios that can be problematic for other allocators:

• Allocations on one thread are freed by a different thread
• Deallocations occur in large batches

Dropping a Box or Vec from another thread than the one it's created from involves an atomic operation.
It works, but for Program, we have one important fact we can exploit - almost all Box and Vec in one Program are allocated by the parser, and within a single thread. It means, we actually don't have to pay for all atomic operations in nested fields.

We can have a field with a type like oneshot::Sender<ast::Program> along with the ast::Program, and send it back to the thread the Program is created from.

Obviously, this will make the code more complex, but it will save enormous amount of atomic operations.