[kronberger-droid]

Summary

Painter tests so far assert on the escape bytes the painter emits, or replay them through the small hand-written emulator in painter.rs (replay).
Bytes say little about what reaches the screen, and they break on any reordering that is correct.
Worse, the expected rows usually come from the same layout math under test, so the painter ends up being its own oracle.

This adds vt100 as a dev-dependency and a thin TestScreen wrapper over it (src/painting/test_screen.rs).
A test paints into the capture writer, feeds the bytes through TestScreen, and asserts on rows, the cursor, cursor visibility and how many rows scrolled into history.
Tests only talk to TestScreen, so the emulator can be swapped by changing that one file.

src/painting/painter/screen_tests.rs has the first tests on it.
They build PromptLines through PromptLines::new the way the engine does, so a new field on PromptLines does not reach them.

No public API change, everything is #[cfg(test)].
The only non-test change is the vt100 dev-dependency in Cargo.toml and Cargo.lock (MSRV 1.70, below ours).

Before

Painter tests check byte strings such as \x1b[?25h\x1b[J\r\n, or a screen rebuilt by replay, which only models the sequences the painter happened to use when it was written.

After

Painter tests can state what the user sees:

session.paint(&FixedPrompt::new("> ", ""), "hello", "");

assert_eq!(session.screen.row(2), "> hello");
assert_eq!(session.screen.cursor(), (7, 2));
assert_eq!(session.screen.scrolled(), 0);

Additional notes

Neither is filed yet, i will open issues for them and fix them separately.

Where vt100 is just one opinion: it keeps a deferred wrap across DECSC/DECRC and wraps a wide grapheme early with one column left.
The module doc says so, tests should assert that the painter stays out of such cases rather than on how vt100 resolves them.

Out of scope:

• Porting the existing replay based tests and removing replay, follow up.
• Fixing the two recorded bugs.