So far, the same algorithm is used to compute an (upper bound on) the maximum witness size for both P2WSH and P2TR miniscript. That's unfortunate, because it means fee estimations for P2TR miniscript will miss out on the generic savings brought by P2TR witnesses (smaller signatures and public keys, specifically).
Fix this by making the algorithm use script context specification calculations, and add tests for it. Also included is a cleanup for the tests to avoid mutable globals, as I found it hard to reason about what exactly was being tested.