fix(graph): return outgoing edges in document order

[michaelpporter]

Refs #790

petgraph prepends each new edge to a node's adjacency list, so get_outgoing_edges / get_filtered_outgoing_edges (and the grouped variant, which wraps them) returned a note's edges newest-first. Anything that takes the first match ("Jump to first neighbour", PrevNextView, walk_to_roots) picked the last field in the note. Both queries now reverse to insertion order, which is the order the builders read fields from the note.

traverse_note's iterative DFS emitted children in reverse link order, which the old reversed graph order had been cancelling out. It now pushes children reversed, so it keeps link order. Side effect: when two children both link to the same note, the earlier child now claims it.

Tests:

• 3 new Rust tests using the issue's 2026-W38 example (flat, filtered, grouped). They failed before the fix with exactly reversed order.
• 1 new traverse_note test.
• Dev vault: 14 of 15 notes with repeated explicit field types now match document order. The remaining one mixes typed_link and list_note edges; each builder is in document order, and the builders run one after the other.