fix: let a mind map close back up when a node is deleted (#513)

[vibhavkatre]

Deleting two of three child nodes left the last one where the other two had pushed it, instead of moving back toward its parent.

Cause

Adding a node shoves its siblings aside and re-flows the tree — the auto-tidy of #273. Deleting one had no matching call, so the survivors kept the spacing of a tree they were no longer in.

Verified before building it

The issue asked to confirm the scope, since the report also asked for level-spacing logic. Reflowing on delete is enough: mindmapLayoutPatches already spaces a level against the real measured boxes (#427 item 8), so no layout work was needed — the layout was simply never asked to run. The added test pins the numbers: the survivor sat 65.5px off the root's centre line before and after the delete; it now moves in.

The fix

One settle step that every delete path routes through, so it does not matter how a node leaves:

Path	Reached by
deleteMindmapSubtrees	Delete on a selected mind-map node
removeShapes	plain shape delete
removeSelectionOrIds	mixed selection delete
removeWhiteboardSelection	delete on a unified canvas

Two details the issue called out, both handled:

• The anchors are read before the removal. A node that is already gone cannot name the tree it was in, so the surviving parents of the departing nodes are collected first. A deleted root contributes no anchor, which is correct — its whole tree left with it, so there is nothing to close up.
• Several nodes reflow once, not per node. reflowTree settles a whole tree from any one member, so N members of one tree would be N identical re-flows. Trees already settled are tracked by their component set. fitMindmapNodes had the same per-id loop and now shares the deduped helper.

The settle runs inside the caller's commit(), so delete-and-settle is one undo step.

Also fixed in passing

deleteMindmapSubtrees filtered shapes and connectors but never state.selection, so deleted ids stayed selected. The shared removal it now calls clears them. Covered by a test.

Tests

Six, in useDiagramStore.freefloating.test.js. Three of them fail on main:

• the gap left by deleted siblings closes (the regression)
• a plain removeShapes settles too, not only the mind-map Delete key (the regression, via the other path)
• deleted ids leave the selection (the incidental bug)
• delete and settle are one undo step, restoring the survivor's exact position
• an independent second tree on the same canvas does not move (Bug: New Mind Maps and Flowcharts are appended to existing diagrams instead of creating separate ones #48)
• deleting a whole tree by its root leaves nothing behind

Full suite green: 1458 tests.

Closes #513

🤖 Generated with Claude Code