perf(datatree): cache module refs and flatten child recursion

[lukecotter]

Summary

Reduce allocation in data-tree child traversal. Behaviour-preserving.

Changes

• Cache the filter/sort module references and the dataTreeFilter/dataTreeSort option flags at initialize() instead of re-resolving them on every traversal.
• getChildren recurses into a single shared accumulator (_collectChildren) so each descendant is pushed exactly once, instead of allocating and re-copying an array at every level of the tree.

Benchmark (Node)

• getChildren over ~250k tree nodes (depth 4, fanout 6): +38% (1.6×)

Tests

Full unit suite green; DataTree spec updated to reflect that module references are now read at initialize time (realistic mock table).