fix: keep restore TUI within terminal bounds (closes #125)

[babarot]

WHAT

Adjust the restore TUI to follow the actual terminal size on WindowSizeMsg. The list height now shrinks to fit, the help row is told the terminal width so bubbles can truncate it, and width-only resize messages no longer collapse the list.

WHY

Reported in #125: in small terminals the gomi --restore view broke. The list items were pushed off the top because the list kept its default 30-row height, and the short-help row overflowed the terminal width and wrapped mid-word.

The WindowSizeMsg handler in internal/ui/update.go only called m.list.SetWidth(msg.Width). The height was never updated and m.help.Width was left at 0, so the help model could not truncate its output.

A naive fix (always derive list height from msg.Height) introduces a regression: the confirm-view cancel paths in updateConfirmView send a synthetic tea.WindowSizeMsg{Width: m.list.Width()} with Height == 0, which would collapse the list to 1 row every time the user dismisses the confirm dialog. The fix therefore guards each dimension independently.

HOW

internal/ui/update.go: in the tea.WindowSizeMsg branch, update width and height only when the corresponding field is positive. The list is sized via SetWidth / SetHeight, reserving 3 rows and 4 columns for the help block that view.go renders below the list with lipgloss.Margin(1, 2). The reserved values are declared as named constants with a comment that points back to the layout in view.go so the magic numbers are traceable.

internal/ui/update_test.go: add regression tests covering the small-terminal scenarios. TestUpdate_WindowSizeMsg_ShrinksListHeight and TestUpdate_WindowSizeMsg_SetsHelpWidth assert the list and help shrink to fit. TestView_FitsSmallTerminal is an end-to-end check that Model.View() does not exceed the requested terminal width and height. TestUpdate_WindowSizeMsg_HeightZero guards against the confirm-cancel synthetic message: after a real resize, a width-only message must not change the list height.

The patch attached to #125 by @u1f992 was used as the starting point; the height-zero guard and accompanying test were added on top to avoid the confirm-cancel regression.

[github-actions]

Code Metrics Report

	main (4f189da)	#127 (a6d1395)	+/-
Coverage	71.2%	71.2%	0.0%
Code to Test Ratio	1:1.5	1:1.5	+0.0
Test Execution Time	28s	6s	-22s

Details

  |                     | main (4f189da) | #127 (a6d1395) | +/-  |
  |---------------------|----------------|----------------|------|
  | Coverage            |          71.2% |          71.2% | 0.0% |
  |   Files             |             24 |             24 |    0 |
  |   Lines             |           1101 |           1101 |    0 |
  |   Covered           |            784 |            784 |    0 |
+ | Code to Test Ratio  |          1:1.5 |          1:1.5 | +0.0 |
  |   Code              |           2739 |           2739 |    0 |
+ |   Test              |           4216 |           4280 |  +64 |
+ | Test Execution Time |            28s |             6s | -22s |

---

Reported by octocov