benchmarks: add end-to-end scroll and caret movement suites

[tig]

Summary

• Adds EndToEndScrollBenchmarks and CaretMovementBenchmarks — full input→render pipeline benchmarks using real key injection, layout, and draw cycles.
• Adds EditorHarness — a lightweight sync wrapper around Application.Create + ANSI driver + Editor for BenchmarkDotNet.

These complement the isolated-layer benchmarks from #50 by measuring what the user actually experiences: keypress → caret move → viewport scroll → full redraw.

EndToEndScrollBenchmarks

Benchmark	What it does
Arrow ↓ to bottom	CursorDown from line 0 until caret reaches last line
Arrow ↑ to top	CursorUp from last line back to line 0
Arrow → to end	CursorRight char-by-char until caret reaches document end (100-line subset)
Arrow ← to start	CursorLeft char-by-char back to offset 0 (100-line subset)
PageDown to bottom	PageDown until caret reaches last line
PageUp to top	PageUp from bottom back to line 0

CaretMovementBenchmarks

Benchmark	What it does
Caret ↓ with scroll counting	Arrow down through 1K lines, counting viewport scrolls at boundary
Caret ↑ with scroll counting	Arrow up from bottom, counting viewport scrolls
Caret → horizontal scroll	Right through 500-char line in 40-col viewport, counting horizontal scrolls
Caret ← horizontal scroll	Left from end of 500-char line, counting horizontal scrolls
Home/End oscillation	500 alternating Home/End on a long line — worst-case horizontal viewport churn
Ctrl+Home/End oscillation	100 alternating Ctrl+Home/End — worst-case vertical viewport churn

Sample results (ShortRun, Apple M4 Max, 1K-line doc)

Benchmark	Mean	Allocated
Caret ↓ (1K lines)	1,305 ms	5.2 GB
Caret ↑ (1K lines)	1,320 ms	5.2 GB
Caret → horizontal	129 ms	570 MB
Caret ← horizontal	132 ms	571 MB
Home/End oscillation	85 ms	363 MB
Ctrl+Home/End oscillation	130 ms	533 MB

The vertical scroll numbers (1.3s for 1K lines, 5.2 GB allocated) make the case for visual-line caching (#49) very clearly — each of the 1K arrow-down steps rebuilds all 24 visible lines from scratch.

Running

dotnet run --project benchmarks/Terminal.Gui.Editor.Benchmarks -c Release -- --filter "*EndToEnd*"
dotnet run --project benchmarks/Terminal.Gui.Editor.Benchmarks -c Release -- --filter "*CaretMovement*"

Test plan

• dotnet build Terminal.Gui.Text.slnx succeeds
• dotnet format --verify-no-changes passes
• CaretMovementBenchmarks dry run completes successfully
• EndToEndScrollBenchmarks dry run in progress (arrow ↓/↑ at 500 and 5K lines validated)
• All existing tests pass (53 editor + 85 integration)

🤖 Generated with Claude Code