[tkc]

The session list dropped every row that did not fit: with fourteen sessions
open the last ones were simply not drawn, and a jump key could select a row
nobody could see.

Stacked on #44 — it rewrote sidebar_layout, which is exactly where this
lands. GitHub will retarget this to main once #44 merges.

What it does

• The list starts at a row offset and scrolls. The wheel over the left pane
  moves it; it used to fall through to the selected terminal's scrollback,
  which is not what the pointer was on. When the cross-session hits are
  showing, the wheel moves those instead.
• Switching sessions brings the selected row back into view.

Rows have different heights (the title line and the branch line are both
optional), so the offset counts tree rows and the placement walks the
measured heights. Three pure functions carry it, each with its own test:
sidebar_rows (what fits from an offset), sidebar_max_offset (the row that
puts the last one on the bottom edge — the wheel stops there rather than
scrolling into blank space), sidebar_follow (raise the offset until the
selected row is inside).

Following only happens when the selected row changes, not every frame.
Every frame would undo the wheel the moment the hand left it. The clamp to the
last usable offset does run every frame, since closing a session lowers it.

Two things the eye check caught

The visual gate is TERMIT_PROBE_OVERLAY=many cargo run --release -- --probe out.png, added here: 21 sessions, the last one selected, the list scrolled.

1. The jump labels were wrong. They were taken from the position inside
   the window, so after scrolling ⌘1 sat on a row that ⌘1 does not select
   — against what the README promises. Now taken from the tree row.
2. Drag-to-reorder has two coordinate systems that used to coincide.
   sidebar_boundary returns a position in the window; drop_row takes tree
   rows. With a non-zero offset a dragged row would land that many places
   away. sidebar_boundary now returns the tree row and sidebar_boundary_y
   converts back.

Also

Rows outside the window now reconcile shown_working. Leaving them stale made
wants_redraw ask for a full redraw on every byte a session nobody could see
printed — pre-existing for the cut-off rows, but this change owns them now.

Gates

cargo fmt --check clean, clippy --all-targets -D warnings silent,
308 passed; 0 failed; 2 ignored (7 new).

Not in this PR

No hint of how many rows are hidden above or below. The labels give it away
(⌘A at the top means you are nine rows down), and a counter is a separate
drawing change.

🤖 Generated with Claude Code