fix(plustek): restore Prescan crop x-mirroring for mirrored sensors

[TobbyTravel]

Summary

Fixes the Plustek Prescan crop being horizontally displaced ("correct crop size, offset one way") on mirror_x devices (OpticFilm 8200i SE / 8100 V2).

Root cause — a regression from #958

• pyopticfilm flips the image left–right in assemble() (mirror_x) because the sensor reads right-to-left. The Prescan the user frames on is therefore the mirror-corrected image.
• But NegPy passed the crop window unchanged into TA/hardware space. The hardware crops by sensor x (pre-mirror) while the user drew by display x (post-mirror), so the returned ROI was reflected → right dimensions, displaced x.

The mirror-aware conversion existed and was correct since the Plustek backend shipped (#783) — crop_to_scan_window(..., mirror_x=...) with the x-reflection (x1,y1,x2,y2) → (1-x2, y1, 1-x1, y2). PR #958 ("feat(plustek): upgrade to pyopticfilm 1.1.2") silently replaced it with a plain clamp_scan_area at all three call sites and deleted the capability plumbing. This PR restores that behavior.

Changes

• negpy/infrastructure/scanners/params.py: restore crop_to_scan_window(crop, *, mirror_x) — self-inverse x-reflection.
• negpy/infrastructure/scanners/base.py + plustek_backend.py: restore prescan_mirror_x capability, wired from model.mirror_x.
• negpy/desktop/view/widgets/prescan_dialog.py: apply the conversion at all three crop↔window points (default-crop display, window change, accept).
• Tests: x-reflection, passthrough, self-inverse; prescan_mirror_x surfaced on 8100 V2 caps.

Validation

• Full suite: 4882 passed, 12 skipped.
• Hardware: verified on OpticFilm 8100 V2 — a Prescan crop now lands exactly where drawn.

Scope

Plustek-backend only. PrescanCropDialog is gated on capabilities.prescan, which only Plustek sets; SANE/PieUSB use their own window handling and are unaffected. (Confirmed during investigation.)

[TobbyTravel]

Paired with the GL128 first-scan prime (jboneng/pyopticfilm#16) on the OpticFilm 8100 (V2): Prescan crops land exactly where drawn, stable across single-pass, Multi-Exposure and settings changes within one device session.

For review, the pairing has a versioning wrinkle: this PR is fully green against the current pyopticfilm==1.2.0 release (lock) and mergeable as-is. But upstream pyopticfilm main has since renamed bringup.is_opticfilm_8200i_se → is_gl128_opticfilm (commit 92d513d, post-1.2.0, and widened to accept the 8100 V2 pid 0x1824 alongside the 8200i SE 0x1825). The next pyopticfilm release after the #16 merge will necessarily carry that rename, at which point the Plustek backend's imports break. Suggested follow-up (small, could also be folded in pre-merge): bump the plustek extra floor to >=<next release>, uv lock, and adopt the new symbol name — the rename is strictly more correct (it matches models by asic == GL128, with the SE pid kept for fallback), so no behaviour change for 8200i SE.