os/v3/puppy: a console-TUI dialog loses its question before its hint - #1028

#1028 ready for review at 1c50d986 (rebased on main bdcd71e6; main has not touched os/v3/puppy/console-tui/ since the base, so this is conflict-free).

The one measurement to check: a pty whose pane is exactly 40x11, key 1. Before this branch that screen drew the key-confirm question with no hint line at all (y mints one and restarts Lemonade, n cancels), and y still minted; the same size now keeps the hint whole and gives up question rows instead. 80x25 and 240x67 are byte-identical to the committed snapshots.

Independent of #1026 (f245bb9e): different hunks, different block, either merge order is safe - both touch the README's test-count line with the same new value, so nothing there to resolve.

Head moved to 7ba4e7f3 for one reason: a comment in the diff said the reboot dialog was Reboot the box now? and the dialog's own words are Reboot the appliance now?. Content otherwise identical (git diff 1c50d986 7ba4e7f3 is that comment, 2 lines in / 2 lines out).

Artifact hash correction, so nobody chases a ghost: the tree of this head builds 42e11547b870428c, 1,556,736 B - not the c046551247f07569 (1,556,728 B) the earlier comments quote, which is the pre-comment-fix tree. Measured, not assumed: both trees were built twice and each pair is byte-identical, so a comment-only change really does move this binary by 8 bytes (this repo already knows rustc embeds source locations - #1014's state file has the same note for 274811c2… vs 0edbe573…). Behaviourally identical; 42e11547… is what a rebuild from this head produces.

CI is re-running on the new head.

Updated to main (8c1eaea3): head 12907048, git diff 7ba4e7f3 HEAD -- os/v3/puppy/console-tui is empty, cargo test 47/47 on the merged tree. The reviewed diff has not moved; the artifact line in the commit message still holds (42e11547b870428c is this tree).

Head deb9a061 - placement only, no behaviour change, and it exists because of the note in the review: "whichever lands second should re-run cargo test after its merge".

Measured first: test-merging #1026 (39da6108) into this branch conflicted on both files (README.md, src/ui.rs) - the two PRs appended their tests at the same end of the module and inserted their README paragraph after the same anchor. So the modal test now sits above the_menu_rule_gives_way_before_an_action_does and the README paragraph above the MENU-rule paragraph. After that, both merge orders merge cleanly and the combined tree is 48/48:

#1026 -> #1028   Automatic merge went well; stopped before committing   48 passed
#1028 -> #1026   Automatic merge went well; stopped before committing   48 passed

One bookkeeping nit, recorded rather than hidden: both PRs set the README's test count to 47, so after both land the line says 47 while the suite is 48. Whichever merges second can bump it, or tell me and I will - it is a one-word README change and I did not want to put a third push on a PR that is already READY for a count.

The reviewed diff of the code is unchanged: git diff 12907048 deb9a061 -- os/v3/puppy/console-tui/src/ui.rs is the test moving, nothing else; cargo test 47/47 on this branch alone.

#1026 merged as 9315118d (17:53 AST), so this branch has been merged onto it: head 38b8e1d8, cargo test 48/48 (46 + one test from each PR), and the README count now says 48 - the one nit from the review is closed.

Two consequences worth stating because my earlier comments quote the old numbers:

• Artifact hash moves again, and this is the one a rebuild from this head produces: 5261cf561e4eaea4, 1,567,800 B (the previous 42e11547… was this branch before os/v3/puppy: a console-TUI message costs the legend a word, not an action #1026's content arrived, and c0465512… before that). The binary grows because it now carries both layout fixes. build-static.sh prints the hash on every build, so a cut can re-derive it rather than trust this line.
• The merge-order work in deb9a061 is now only of interest as bookkeeping: os/v3/puppy: a console-TUI message costs the legend a word, not an action #1026 is in main, so this PR merges against it directly. The test-merge measurements in that comment (#1026 -> #1028 clean, 48/48) are what this head contains.

Everything else stands: the reviewed modal diff is unchanged (git diff deb9a061 38b8e1d8 -- os/v3/puppy/console-tui/src/ui.rs is empty - the merge and the README count are the only changes), 80x25/240x67 frames byte-identical, every gate string unchanged.

(A previous version of this comment was mangled by shell backticks in transit; this one is the correction.)

Re-verdict requested at 38b8e1d8 - the head has moved twice since the READY at 12907048: deb9a061 (placement-only: the modal test moved above the MENU-rule test and the README paragraph above the MENU-rule paragraph, so #1026 and this PR stop sharing merge context) and 38b8e1d8 (main merged, i.e. #1026's own change, plus the README count 47 -> 48).

For a fast re-check, the two things that could have moved: git diff 12907048 38b8e1d8 -- os/v3/puppy/console-tui/src/ui.rs is a test moving plus #1026's footer work arriving through the merge (no line of the modal fix changed), and cargo test --offline is 48/48 on this head. 80x25/240x67 frames byte-identical; gate strings unchanged.

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

let (title, text, hint) = modal_box(m);
let lines: Vec<String> = std::iter::once(title.clone())
    .chain(text.iter().cloned())
    .chain(std::iter::once(hint.clone()))
    .collect();
// the box is the longest line plus its own padding: sizing it to the longest line and then wrapping
// inside it made every hint wrap for no reason (measured: "any other key / cancels")
let width = (lines.iter().map(|l| l.chars().count()).max().unwrap_or(10) + 4)
    .min(w.saturating_sub(4))
    .max(27);
let wrap = |s: &str| -> Vec<Line<'static>> {
    wrap_words(s, width.saturating_sub(4))
        .into_iter()
        .map(|l| styled(l, Some(Color::White), false))
        .collect()
};
let mut title_rows = wrap(&title);
let mut text_rows: Vec<Line> = text.iter().flat_map(|l| wrap(l)).collect();
let hint_rows = wrap(&hint);
// The hint is the only row that says how to answer the question, so it is never the row that goes: a frame
// too short for the whole box loses the explanation first, whole rows at a time, and then the title. Before
// this the box was clamped to the frame and ratatui clipped the END - at 40x11 the reboot confirm drew
// `Reboot the appliance now?` with `y reboots the box, n cancels` off the screen while `y` still rebooted.
let room = (area.height as usize).saturating_sub(2);
while title_rows.len() + text_rows.len() + hint_rows.len() > room && !text_rows.is_empty() {
    text_rows.pop();
}
while title_rows.len() + hint_rows.len() > room && !title_rows.is_empty() {
    title_rows.pop();
}
let body: Vec<Line> = title_rows
    .into_iter()
    .chain(text_rows)
    .chain(hint_rows)
    .collect();
let bh = (body.len() as u16 + 2).min(area.height);

if let Some(m) = &a.modal {