feat(helper-lines): add smart alignment and paint snap opt-out

[saif-ellafi]

Description

Hello, this is a bit of an experimental, AI-Assisted branch that improves snapping with helperLines.
The motivation is to soften the magnetic force field without having to disable edge snapping (as previous PR did) to improve the user experience.

However, Smart Snapping is quite opinionated, and I put it on a way to be optional, but feel free to decide what you think out of it. Tested manually.

This adds opt-in smart alignment so same-kind layers snap matching edges (left↔left, etc.), mixed types stay center↔center, and only nearby / shared-axis neighbors participate. Canvas midlines snap centers only when smart is on. Custom guides stay wildcards.

Snapping is measured from the pointer’s intended position, so both axes can catch and still be left (diagonal L drags don’t get stuck).

Also adds enablePaintLayerSnapping (default true). When false, paint is ignored as a snap target and does not snap while dragging. This is because painting layers are free form weird shaped things, I found it annoying to snap to a drawing layer, I think it goes well together.

Defaults keep current behavior: enableSmartAlignment: false, enablePaintLayerSnapping: true.

Related Issue: Closes #

Type of Change

• ✨ New feature (non-breaking change which adds functionality)
• 🛠️ Bug fix (non-breaking change which fixes an issue)
• ❌ Breaking change (fix or feature that would cause existing functionality to change)
• 🧹 Code refactor
• ✅ Build configuration change
• 📝 Documentation
• 🗑️ Chore

[hm21]

Thanks a lot for this PR — really nice work. The problem you describe is real: on a dense canvas the old magnet field gets sticky, and disabling edge snapping entirely was too blunt a fix. Measuring the snap from the pointer's intended position instead of the glued position is the right call, and it's what makes diagonal drags catch both axes without trapping the layer.

I reviewed it carefully and confirmed the default path (enableSmartAlignment: false, enablePaintLayerSnapping: true) is behaviour-preserving — addTarget, _anchorForTarget, _heldAlignKey and _canvasSnapAnchor all collapse to the previous logic when the flag is off. The one delta, releasing held targets at max(snapThreshold, releaseThreshold), is a no-op at normal zoom and fixes a case when zoomed far out where a target could be released and re-trapped in the same frame. Good catch.

I pushed three small follow-ups on top of your branch rather than sending you back and forth:

• Tunable reach — _smartAlignmentNeighborLimit (3) and _smartAlignmentSharedMin (2) are now HelperLineConfigs.smartAlignmentNeighborLimit / smartAlignmentSharedAxisMin. The feature is opinionated by design, so apps should be able to dial it in rather than take our numbers.
• Shared axes snap to a real edge — a cluster previously snapped to the average of its positions, which can sit up to snapThreshold off every actual layer edge, so the guide line didn't quite overlay anything. It now picks the real edge closest to the cluster mean.
• Neighbour distances measured once instead of recomputing _snapLayerCenter inside the sort comparator on every drag frame.

Plus the CHANGELOG entries and tests for the new fields. Merging now — thanks again!