Comparison with free list

[CAD97]

Rowan currently uses a handrolled thread local free list to avoid allocating and deallocating syntax node payloads a large amount.

I'm currently writing an experimental rewrite from the ground up with a more aggressive eye for optimization. I'd love to use remem over an ad-hoc solution.

For that purpose, it'd be great to have some comparisons of remem against a simple free list, both thread-local (Rc) and shared (Arc).