feat(scenario): add ScreenBuffer::occupied_rows() for flood assertions

[Roger-luo]

Summary

Adds ScreenBuffer::occupied_rows() to the scenario crate: the number of screen rows containing at least one non-space character.

This is a reusable primitive for asserting that transient terminal output — a status line, a progress/log tail, an in-place spinner — stays bounded and doesn't flood the screen. Today callers hand-roll this with visible_screen().iter().filter(|l| !l.is_empty()).count() or screen[N..].iter().all(|l| l.is_empty()); occupied_rows() names the intent and lives next to the rest of the ScreenBuffer API.

Motivating use case: a downstream CLI (autotune) renders a dimmed rolling "tail" of subprocess output via cursor-up + erase escape sequences. A regression there let the tail flood the screen when lines wrapped. occupied_rows() lets a PTY test assert the rendered screen never fills up — the kind of "no-flood" check the raw output stream can't express (only the vte-rendered screen reflects the erases).

Changes

• ScreenBuffer::occupied_rows(&self) -> usize — counts rows with non-space content (post-vte, so cleared/erased rows don't count).
• 4 unit tests: non-blank rows, blank screen, rows cleared by ESC[2J, and sparse content (only rows with text counted).

Test Plan

• cargo nextest run -p scenario — full suite passes (incl. 4 new occupied_rows tests)
• cargo fmt -p scenario clean
• cargo clippy -p scenario --all-targets -- -D warnings clean

🤖 Generated with Claude Code