Add Harness Evolver to Benchmark/Evaluator

[raphaelchristi]

Summary

Adds Harness Evolver to the Benchmark/Evaluator section.

What it is

LangSmith-native autonomous agent optimization. Multi-agent proposers evolve prompts, routing, tools, and architecture in isolated git worktrees with LLM-as-judge evaluation and regression guards.

• npm: npx harness-evolver@latest

[Jenqyang]

Decision: Not ready to merge yet.

Reason:

1. The candidate may fit, but the current line does not clearly justify why this belongs in Benchmark/Evaluator rather than a tooling/optimization bucket.
2. The README makes LangSmith a central runtime dependency, so under the updated standard we need that commercial dependency boundary to be framed more clearly.
3. The wording should be tightened to a more neutral technical description.

Next step: Please clarify the section fit, make the LangSmith dependency explicit in neutral terms, and tighten the line, then we can re-review.