fix: keep a zz centering when opening a comment near EOF

[davinkevin]

Problem

zz on a line near the end of a file centers the cursor and leaves blank rows below the content. Opening a comment on that line snaps the view back, so the line being reviewed jumps to the bottom of the viewport. Esc does not restore the centering.

Repro

1. Open any diff and move the cursor near the end of a file — far enough that zz leaves blank rows below the content.
2. Press ZZ to center the line.
3. Press c to open a comment on it.

The line drops from the middle of the viewport to near the bottom. It happens in both the unified and side-by-side views.

Cause

Two places disagree on the maximum scroll offset.

App::max_scroll_offset (src/app/navigation.rs) explicitly allows blank space below the content, which is what makes zz work at EOF:

/// This permits empty space below content (e.g. when centering the cursor near EOF)
pub fn max_scroll_offset(&self) -> usize {
    self.total_lines().saturating_sub(1)
}

scroll_comment_input_into_view (src/ui/diff_view.rs) clamped with a stricter "no blank space below" rule:

let max_scroll = total_lines.saturating_sub(viewport_height);

That clamp runs on every frame an input box is rendered (diff_unified.rs, diff_side_by_side.rs). The early return only checks whether a box exists, so the clamp fired even when the box was already fully visible and no adjustment was needed.

Worked example — viewport 40 rows, 100-line document, zz on line 98:

step	scroll_offset	screen row of line 98
after zz	78 (18 blank rows below)	20 — centered
comment opened (4-line box, 104 lines total)	clamped to 104 − 40 = 64	34 — near the bottom

Fix

Clamp to total_lines - 1 so the render path and the app agree. The branches that pull the box into view when it is above, below, or taller than the viewport are untouched, so the box still follows the text cursor while typing.

Tests

• should_keep_cursor_centered_when_opening_comment_near_eof — the scenario above; scroll_offset stays at 78.
• should_clamp_to_last_line_when_box_sits_past_the_content — covers the changed clamp directly.
• should_not_scroll_past_end_of_content asserted 90, but that value came from the box-above-viewport branch, not the clamp — the clamp never ran in that case. It now asserts the 95 that branch produces, and the new test above pins the clamp.

cargo test (1363 passed), cargo clippy --all-targets -D warnings, and cargo fmt --check are clean. Also verified interactively with a release build: centering survives at EOF, and the input box still auto-scrolls into view at the top and middle of a file, in both diff layouts.

No issue filed for this — I could not find one open that tracks it (closest is #76, different code path; #11 is likely where the strict clamp came from).

[agavra]

nice, this is subtly improved.