Efficient Trace Lowering

Right now, we lower each AST in a trace separately with an arg substitution. This makes us repeat lowering an edge contracting the same protocol AST since each time its a different arg substitution. If we lowering each protocol once, then edge contract, and then make copies for each argument substitution it'll be more efficient.

Locate the trace-lowering entry point and the code that contracts protocol AST edges. Compare the current per-AST lowering with argument substitution against the proposed flow of lowering each protocol once, contracting edges, and copying for substitutions. Done means equivalent traces are produced without repeating protocol lowering and the relevant tests or benchmarks pass.

Written by the indexing model from the issue text.