[kvlonge]

Why

A selected Python task or asset only ran after every declared SQL dependency reported a result in the same build. A dependency outside the run never reports one, so sqb build --select task:x skipped the task, and a task on an unmanaged source was skipped on every build. SQL models already build against existing upstream relations when those upstreams are not selected.

Changes

• The read-side Python tracker gates a node only on SQL resources scheduled in this run (models, seeds and source loads in the plan's execution order, plus pre-SQL source loads). Dependencies outside the run are read as they exist.
• A skipped upstream model now skips its Python dependents, matching how the build blocks downstream models; seeds and source loads still gate on failure only.
• sqb check no longer fakes success results for dependencies.
• Python nodes overview documents the selection rule.

Verification

• New DuckDB E2E test_python_node_selection.py (11 cases; 7 fail on the old code) covering tasks and assets selected alone, unmanaged sources, waiting on and skipping after a selected upstream, and a skipped upstream model.
• The deferral E2E now asserts the task-only run reads the deferred source.
• make test, make check-ci and the hygiene scan pass; one independent review.