JMH benchmark suite for render + layout

[rorygraves]

Background

No benchmarks exist. Any perf claim about Layout v2, diff rendering, or virtual ListView is anecdotal. Regressions are invisible.

Goal

A JMH-based benchmark project with microbenchmarks for critical paths, runnable locally and in CI (gated, not on every commit).

Scope

1. New sbt submodule modules/termflow-bench using sbt-jmh.
2. Benchmarks for:
   • Layout.measure on representative trees.
   • Layout.resolve → VNode.
   • AnsiRenderer.diff on controlled before/after framebuffers.
   • End-to-end frame cost on SineWaveApp.
3. Baseline numbers checked into docs/PERFORMANCE.md.
4. CI job: sbt bench/Jmh/run -i 1 -wi 1 -f1 smoke-run; full runs manual.

Acceptance

• sbt bench/Jmh/run works locally.
• Baseline numbers captured in a doc.
• CI smoke-run integrated.

Part of Phase 4 — Rendering & perf.

[rorygraves]

Done — implemented via a 6-PR sequence (multidev plan mode):

• Jmh bench suite bench module scaffold #218 scaffold the non-aggregated termflow-bench sbt-jmh module
• Jmh bench suite bench ci smoke workflow #219 dedicated bench.yml smoke workflow (the only CI gate for the excluded module)
• Jmh bench suite bench fixtures and layout #220 Fixtures + LayoutMeasureBench / LayoutResolveBench
• Jmh bench suite bench render buildframe diff #221 BuildFrameBench + AnsiDiffBench
• Jmh bench suite bench sinewave end to end #222 SineWaveFrameBench end-to-end per-frame bench
• Jmh bench suite bench baseline doc #223 docs/PERFORMANCE.md baselines (Apple M4 Max, Corretto 21.0.7, JMH 1.37)

All acceptance criteria met: sbt "bench/Jmh/run" works locally (44 variants), baselines captured, CI smoke-run integrated. Numbers are documented as indicative, not a CI threshold.