Fix three ADF -> Databricks control-flow / codegen bugs

[matthewmoorcroft]

⚠️ Draft — initial port. This change is rebuilt onto the unified base as-is. It predates the engine → sources/adf module restructure, so it still needs adaptation to the new layout (and a green test run) before it is mergeable.

1. ForEach inputs-bridge notebooks now carry the # Databricks notebook source marker so bundle validate accepts the notebook_task.
   • Already fixed on main by the render_bridge_notebook consolidation (the bridge-notebook header fix). No source change here — this PR only adds a guard/characterization test for it. So the source diff shows two behavior changes, not three; a reviewer isn't left hunting for a missing third source fix.
2. A join after an IfCondition uses run_if: NONE_FAILED so a failure in the taken branch propagates (AT_LEAST_ONE_SUCCESS masked it).
3. Synthesised _init_<var> tasks that nothing references are pruned. The keep/drop decision matches structured references (the anchored tasks.<key>.values task-value token and depends_on edges) rather than a bare substring, so a dead _init_x is never kept alive by an unrelated key/path such as _init_x_2 or /Shared/_init_x_report.

Co-authored-by: Isaac

Changes

Tests

• manually tested
• added unit tests
• added integration tests

[matthewmoorcroft]

Closing — this was a port of another author's work and is being withdrawn from the public repo.