Fix braille rendering termination and frame completion

[rustfuture]

Problem

The braille renderer walked floating-point coordinates until exact endpoint equality, could hang on fractional or very large values, could round edge-adjacent points out of bounds, and overflowed while clipping finite maximum-magnitude crossings. CPU labels could be cleared by graph rendering, and completed frames had no uniquely placed flushed marker.

Behavior

• Define empty and zero-size inputs as no-ops and an inclusive-edge single-point policy that clamps right/bottom boundaries to the final dot.
• Clip every finite segment to the logical viewport with overflow-safe intersection math, then rasterize clipped integer cells with direct-y Bresenham.
• Skip NaN/infinite and fully outside segments while drawing crossing segments completely with viewport-bounded work.
• Keep CPU and memory labels outside graph clear rectangles without changing network interval or totals text.
• Write the sole Frame: N marker as the final frame output and flush it immediately.
• Remove the interrupted standalone probe and cover the renderer through killable subprocess timeout tests.

Tests

• cargo fmt --check
• cargo check --locked --all-targets
• cargo clippy --locked --all-targets -- -D warnings
• cargo test --locked (75 passed)
• cargo bench --locked --no-run
• cargo +1.88.0 check --locked --all-targets
• Real PTY runs at 80x24 and 110x50 with GRAINX_REFRESH_INTERVAL_MS=100: frames 1-5 strictly increased; q exited 0; no signals.

Limits

The PTY evidence covers the debug binary on one macOS Apple Silicon host at the two requested terminal sizes. It demonstrates bounded completion and graceful quit, not performance or cross-platform behavior. The temporary PTY driver is intentionally not committed; concise build and run evidence is recorded under demos/.