Make transpiler-performance docs reproducible or rewrite them qualitatively

[justin808]

Problem

docs/transpiler-performance.md currently presents precise benchmark numbers, hardware details, and methodology claims (for example, averages over multiple runs), but this repo does not include:

• a benchmark harness
• raw benchmark outputs
• fixture apps / commands used to produce the numbers
• any reproducible source for the documented figures

That makes the guide hard to trust and impossible to refresh safely.

Suggested fix

Choose one of these directions:

1. Add benchmark scripts, fixture apps, raw results, and a brief reproduction section to the repo.
2. Rewrite the guide as qualitative guidance (tradeoffs, expected directionality, caveats) and remove exact numbers that cannot be reproduced from source control.

Why this matters

Performance claims age quickly. Without reproducible artifacts, the docs risk becoming stale in a way that looks authoritative.