Add a unified Preferences dialog (File > Preferences…)

[PierreRaybaut]

Summary

Introduce a unified Preferences dialog for DataLab-Web, mirroring DataLab Qt's settings system. Accessible from File > Preferences… (consistent with the desktop application).

Motivation

A handful of behaviours are already configurable through ad-hoc UI bits (theme toggle, HDF5 import flags, Results tab auto-switch). As the number of tunable behaviours grows, a single discoverable entry point becomes worthwhile.

When to implement

Implement this once ~8–10 preferences are genuinely worth exposing and wiring to runtime behaviour. Below that threshold, prefer inline controls (theme toggle in the menu bar, checkbox in the HDF5 import dialog, etc.).

Candidates ordered by perceived value:

Option	Category	Effect to wire
Colour mode (auto / light / dark)	general	ThemeProvider.setMode
Auto-switch to Results tab after analysis	proc	App-level side panel switch
Clear workspace before HDF5 import	io	h5_browser_import
Use full HDF5 path as object title	io	import_nodes
Append HDF5 file name to object title	io	import_nodes
Auto-downsampling of large signals	view	Plotly backend threshold
Lock image aspect ratio to 1:1	view	Plotly yaxis.scaleanchor
Eliminate outliers on image import	view	Percentile clip
FFT shift by default	proc	Default param in FFT wrapper
Auto-normalise convolution kernels	proc	Default param in convolution
Console log level	console	console.* wrapper
Keep analysis results in metadata	proc	Metadata purge policy
Pairwise n→1 mode	proc	⚠ requires implementing the dispatch in processor.py first

Design

• Python source of truth: one guidata.dataset.DataSet subclass per category, with a category_id class attribute. The JSON schema is produced via dataset_to_schema_with_values and rendered by the existing DataSetForm.
• Persistence: a single JSON blob in localStorage under datalab-web:preferences, namespaced by category_id. Persist only diffs vs defaults, so future default changes propagate naturally.
• Runtime coupling: a PreferencesProvider loads from localStorage on mount, pushes overrides to Pyodide once runtime.status === "ready", and mirrors general.color_mode to ThemeProvider.setMode (with auto → system mapping).
• Menu entry: File > Preferences… (match DataLab Qt; not Edit).
• Dialog: left-rail tab layout with fixed dimensions (e.g. width: 720, content area height: 320, overflowY: auto) so the action buttons don't reflow when switching tabs. Selected tab styled with var(--accent) + white text, like menu items.
• Buttons: Export… / Import… / Restore defaults / Cancel / OK.
• JSON export: export the full effective state (defaults + overrides + unsaved edits), not the diff — otherwise an untouched configuration produces {} and re-import isn't deterministic across default changes.
• JSON import: validate the root shape, drop unknown categories with a user-visible warning, filter unknown keys server-side.

Acceptance criteria

• At least 8 preferences exposed and verifiably wired to runtime behaviour.
• Each option has a behavioural test (not just storage round-trip): Vitest+RTL for UI-facing effects, pytest for Pyodide-side effects.
• Tests cover JSON import (valid payload, unknown categories, malformed JSON, non-object root, non-object category) and export (untouched config yields a non-empty document; round-trip equals identity).
• No schema field labelled "(takes effect in a future release)" — add a field the day its consumer ships.
• HMR-safe Python state (e.g. try: _PREFS except NameError: …).
• Documented in the user guide.

References

• Qt implementation: datalab/gui/settings.py (DataLab desktop)