[kvlonge]

Why

sqb lineage <model> text output could run for minutes and grow past 20 GB of memory on projects with shared dependencies. The tree renderer only tracked the current path, so every shared model was expanded again on every path reaching it, and output grew exponentially with the number of diamonds.

Changes

• Relation and column-trace tree renderers track shown nodes across the whole render: each node is expanded once, later occurrences print (already shown).
• Both renderers are iterative, so deep graphs cannot hit the recursion limit.

Verification

• New DuckDB E2E over a 14-layer diamond project (downstream, upstream and column trace) asserts each node is expanded at most once and output stays linear; the relation cases fail on the old renderer.
• On a private ~2,000-model project the same command went from >135 s / 21.8 GB (killed) to 2.4 s / 79 MB.
• make test, make check-ci and the hygiene scan pass; one independent review.