fix: plate tui — a real two-column prepare screen, and a summary that wraps

[DLANSAMA]

The last known visual gap in the TUI before 0.5.0.

The prepare screen wasted ~40% of its width

Everything was stacked in one VerticalScroll, so the slice estimate — the thing you press Prepare to get — landed below the fold. The form is now on the left, the results on the right, and both are visible at once.

Below 100 columns two halves would be too narrow for a material label (ABS — strong, needs an enclosure (detected in AMS) needs 59 cells), so the screen keeps the single-column layout there and scrolls a finished — or failed — run into view instead of leaving it off-screen.

The results side is a titled Result box with copy saying what pressing Prepare will put in it. Before, the initial state was an empty half-screen with a disabled Start print… floating in the middle of it, which read as a half-rendered widget. round border, not thick — 0d63378 established that thick renders as solid slabs.

The summary restarted wrapped values in the label column

Rows were built with f"{label:<11}{value}", so a value that wrapped continued at column 0:

Printer     Bambu Lab P1S, 0.4mm
nozzle
Material    PLA · Quality:
standard (0.20mm) · Supports: no

nozzle and standard (0.20mm) … read like field names. They now render through one shared Rich grid (bambu_cli/tui/widgets/summary.py) used by both the prepare screen and the confirmation dialog, so continuations indent under the value.

Every cell is wrapped in rich.text.Text, the same idiom as status_panel.py / ams_panel.py / confirm.py — download/naming.py sanitizes <>:"/\|?* but not brackets, so a [remix] in a Printables file name reaches these sinks and would otherwise be eaten as markup (or raise MarkupError on a [/x] shape).

One disclosed behaviour change: the shared grid's value column uses overflow="fold" where Rich defaults to "ellipsis", so a long file name in the confirm dialog now wraps in full instead of being cut short. That dialog is the only screen that starts a physical print; it should not abbreviate the name of the file it is about to print. Everything else about its rendering is byte-identical.

Scope

Layout and presentation only. Untouched: temp-workdir ownership and cleanup, the _material_touched guard against a late AMS read overriding a manual pick, the AMS detect worker, the settings handoff, the confirm handoff. confirm.py remains the single confirm=True call site.

A guard that had gone vacuous

test_presliced_3mf_shows_material_not_applied_caveat asserted "Material PETG" not in preview — a spelling coupled to the old f-string padding. Once the grid landed it passed unconditionally, including in the exact regression it exists to catch (the TUI claiming a chosen material was applied to a pre-sliced file it never sliced). It now asserts on the material name, and fails when the production path is sabotaged to emit it.

Verification

Screenshots were a required review step, not text assertions — rendered via export_screenshot() → rsvg-convert → PNG and read as images at 80x24, 96x30, 100x30 and 120x34, covering the initial, prepared, error, slice-failure, AMS-detected and bracketed-filename states. Three defects in the first cut were found this way and fixed: the empty results column, the misaligned wrap, and a stair-step right edge on the form groups.

Every new test was sabotage-verified (break the thing, watch it fail, restore).

ruff check            All checks passed!
ruff format --check   88 files already formatted
mypy -p bambu_cli     Success: no issues found in 88 source files
bandit -ll            No issues identified
pytest -m "not live"  1419 passed, 1 deselected — coverage 89.09% (floor 83)
syntax_smoke          ok (185 files)
cli_help_smoke        ok (21 subcommands)
python_compat_smoke   ok
ci_workflow_smoke     ok (21 CLI commands, 88 package modules)
check_layers          Layer boundaries OK (1 allowlisted debt edge)
gen_schemas --check   docs/schemas up to date (26 schemas)
uv build + package_contents_smoke   ok (wheel ships summary.py and styles.tcss)

Not verified locally: the Windows/macOS legs and pip-audit (CI only), and rendering in a real terminal rather than Textual's headless screenshot.