Add live multi-diffraction comparison to Show4DSTEM

[bobleesj]

What problem does it solve?

Comparing 4D-STEM reconstructions requires seeing each dataset's diffraction pattern alongside its virtual image at the same scan position or region. Add compare_dp_mode="all" to the existing Show4DSTEM Multiple view, with shared scan-region Mean/Sum/Max reductions and linked detector interaction.

The comparison grid reuses zoom, pan, reset, scale bars and column controls. Region updates use a latest-pending scheduler with acknowledgements, and dragging a diffraction tile updates the common virtual detector while held. Playback is restricted to Single view; switching to Multiple stops playback. Existing selected/average modes remain available.

What should reviewer(s) do?

• Follow storyboard S4D-21 and the new API example: compare several datasets, move a shared scan ROI, and drag a detector from a comparison diffraction tile.
• Verify shared-region reductions, signed values, live updates, zoom/scale bars, and Single/Multiple playback transitions.
• Review the Python buffer publication and frontend request/acknowledgement lifecycle. This is live-kernel comparison; standalone multi-DP HTML and packed-resident compatibility are not qualified by this PR.

Verification

• Production frontend build and TypeScript check passed.
• 22 focused frontend tests passed across region scheduling, detector interaction and progressive comparison.
• 20 focused Python tests passed across all-diffraction behavior and the HTML export protocol.
• Real JupyterLab/CUDA workflow exercised a 64x64 scan with native 192x192 diffraction patterns and five displayed datasets. All five virtual images changed during a held diffraction drag; no browser exceptions were captured.
• Ten headed playback/view checks passed, including stopping playback on entering Multiple and remaining paused when returning to Single. The actual comparison display was visually inspected.
• The full Python suite did not pass in the current environment. A narrower contract run had 31 passes and two failures in source-string assertions targeting an older QuantEM.GPU backend layout. This draft does not claim baseline equivalence or merge readiness; those failures need isolation before promotion.
• No 120 FPS, physical-phone, or standalone-export qualification is claimed. No experimental data, screenshots, generated notebooks or build bundles are included.

---

This PR follows the
scikit-package workflow for
reproducible scientific software.

[github-actions]

• Delete bobleesj/quantem.widget:sep-10-show4dstem-comparison from its remote if you own it, it is not long-lived, and no other PR uses it.
• Remove the dedicated worktree and local sep-10-show4dstem-comparison branch after confirming the worktree is clean.
• Fetch electronmicroscopy/quantem.widget and update local main before starting new work.