Feature: Add reproducible transform benchmarks #1026

Is your feature request related to a problem? Please describe.

We need a reproducible performance baseline before changing how the transform plugin executes programs. Existing diagnostic benchmarks and prototypes have helped identify overhead, but they are not part of the regular benchmark suite. The current BenchmarkSetMoves measures decoding and direct RootTarget operations; it does not exercise the complete Plugin.Do path, function calls, or local variables.

Program preparation is also assembled separately by Plugin.Start and execution test helpers. Future changes to preparation could reach production and benchmarks differently, making comparisons misleading.

The goal is to establish permanent benchmarks, record the current implementation's performance, and provide a shared preparation entry point. This gives subsequent optimizations a reliable correctness and performance reference, including comparisons with modify on equivalent workloads.