TreeNode depth is likely sometimes incorrect

TreeNode.depth assumes that TreeNodes are not modified after the first time depth is accessed but it likely code violates this assumption.
We should probably instead switch to a scheme where TreeNode depth values are invalidated when the tree structure bellow them changes although this needs to be done with care to avoid excessive computations of depth.

Locate the TreeNode implementation and inspect how depth is cached and how tree structure changes are applied. Trace the mutation paths first, then define completion as keeping depth values correct after structural changes without causing excessive recomputation; add or update regression coverage for those cases.

由索引模型根据 Issue 内容生成。