Consider optimizing folds by not using variadic expansion

Right now, the folds are implemented by unpacking the sequences and then folding recursively on variadic packs (with some unrolling). However, passing many parameters around can be difficult to optimize away, and overload resolution is expensive in terms of compile-time. Instead, we should consider having a unrolled recursive fold that indexes into the original structure.

Start by locating the fold implementations and tracing how sequences are unpacked into variadic packs and recursively folded. Compare that path with the original structure and identify the existing unrolling and overload-resolution behavior; done means a validated optimization that preserves fold behavior while reducing unnecessary parameter passing and compile-time cost.

Written by the indexing model from the issue text.