benchmarks: add baseline scrolling performance suite

[tig]

Summary

• Adds benchmarks/Terminal.Gui.Editor.Benchmarks project with BenchmarkDotNet, covering the three hot-path layers that determine scroll fluency.
• Registers the project in Terminal.Gui.Text.slnx and adds BenchmarkDotNet.Artifacts/ to .gitignore.

Benchmark suites

Suite	What it measures	Parameters
ScrollingBenchmarks	Viewport-sized visual line builds at top/middle/bottom, plus full top→bottom scroll simulation	Document: 100/1K/10K lines, Viewport: 24/50 rows
VisualLineBuildBenchmarks	Per-line build cost across content profiles (short ASCII, long line, tabbed, emoji/ZWJ, mixed CJK+tabs+emoji)	5 content types
DocumentAccessBenchmarks	Rope text extraction and line-tree lookups (sequential vs random access patterns)	Document: 1K/10K/100K lines

Running

# All suites
dotnet run --project benchmarks/Terminal.Gui.Editor.Benchmarks -c Release

# Single suite
dotnet run --project benchmarks/Terminal.Gui.Editor.Benchmarks -c Release -- --filter "*Scrolling*"

Sample results (ShortRun, Apple M4 Max)

VisualLineBuild — per-line cost:

Content	Mean	Allocated
Short ASCII (~40 chars)	2.3 µs	11.6 KB
Long ASCII (~200 chars)	13.4 µs	72.0 KB
Tabbed line (4 tabs)	2.6 µs	13.1 KB
Emoji / ZWJ clusters	2.3 µs	10.4 KB

Scrolling — viewport builds (10K-line document, 50-row viewport):

Scenario	Mean	Allocated
Viewport at top	179 µs	837 KB
Viewport at middle	195 µs	926 KB
Viewport at bottom	201 µs	948 KB
Full scroll top→bottom	42.9 ms	184 MB

The full-scroll number (42 ms for 10K lines) and the per-viewport allocation (~900 KB) confirm that visual-line caching (#49) is the right next optimization.

Test plan

• dotnet build Terminal.Gui.Text.slnx succeeds with benchmark project included
• dotnet format --verify-no-changes passes
• All three benchmark suites run to completion (--job Dry)
• Existing tests unaffected (benchmarks are a separate project, not in test path)

🤖 Generated with Claude Code