Thread the resolved printer through rendering for correct per-model output - #56

Summary

Companion to labelle-org/labelle#156, which validates real calibration data for the DYMO LabelManager Wireless PnP (previously an unvalidated placeholder, ~7.1 px/mm, that produced "squished"/incorrectly-scaled output for that model).

That fix only has an effect if labelle-web actually uses it. Right now, rendering (preview and the print payload) and printing each construct a plain DymoLabeler, ignoring device-specific calibration entirely — so a model with its own margins/resolution renders and prints using generic defaults regardless of which printer is actually selected.

• label_builder.py's render functions now accept an optional resolved device, used via DymoLabeler.for_device() when given. Also passes the labeler's real per-model vertical resolution into RenderContext, so the preview's mm-ruler labels match reality for a model whose cross-tape and along-tape resolutions genuinely differ (previously always assumed one shared PIXELS_PER_MM for both axes).
• printer_service.py's actual print path (print_label, print_bitmap) builds its DymoLabeler the same way, so what's printed always matches the resolved device's own protocol.
• app.py's preview and cut-mark-print paths resolve a device via a new resolve_device_for_render (best-effort: falls back to generic rendering for a virtual printer, unresolved printerId, or a scan failure, rather than turning a preview request into an error the way an actual print correctly would).
• Adds 24mm as a selectable tape size (client + server validation) — physically supported per the LabelManager Wireless PnP's own datasheet, though only for models whose SUPPORTED_TAPE_SIZES_MM actually includes it; selecting it for one that doesn't still fails clearly at print time. A fully correct fix would query the selected device's own supported sizes instead of this static list — out of scope here.

Dependency note

This is written against labelle's current main API (DymoLabeler.for_device()). labelle#156 targets develop instead (per a maintainer's standing preference for device-specific work), whose DymoLabeler API differs meaningfully (no for_device(), height_px → get_label_height_px(), preview engines take the labeler instance directly, etc.) — so depending on which branch labelle-org ends up releasing the fix from, this PR may need adjusting to match. Flagging that now rather than have it be a surprise in review.

Test plan

• Full server test suite: 300/300 passing
• Exercised end-to-end against real hardware: correct preview mm-labels, correct print positioning/sizing on 24mm tape
• ruff check clean (pre-existing findings elsewhere in the codebase left untouched, no config enforces them here)

🤖 Generated with Claude Code