Parse Switch case/default activities (fixes undercounted inventory & missing lineage edges)

[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.

📦 Merge order (3-PR set): This PR is independent — it branches off main and depends on nothing else; merge it any time. The other two (the discover-lineage PR lineage → the discover-insights PR insights) are a stack that benefits from this fix but does not require it. Files overlap with the discover-lineage PR/the discover-insights PR only in disjoint regions, so at most a trivial rebase for whichever merges second.

---

Fixes #31

Problem

parse_activity in src/flowx/parser/adf_loader.py descends into ifTrueActivities/ifFalseActivities (IfCondition) and activities (ForEach/Until), but never cases[].activities or defaultActivities on a Switch. So a Switch's nested activities never entered the AdfActivity AST, and every AST consumer inherited the blindness:

• Inventory (_classify_activities) undercounted activity_count / coverage_pct.
• Lineage (_walk_activities) dropped every ExecutePipeline and data edge nested inside a Switch case.

Why this is a bug, not a design choice

Verified via git archaeology and the repo's own docs:

• Switch case-parsing never existed in the parser across its entire history (git log --all -S 'defaultActivities' / -S '"cases"' return only this commit) — nothing was removed "for a reason."
• The Switch translator (translator/activity_translators/switch.py) has read cases[].activities + defaultActivities since the initial release — one half of the tool always understood the shape; the parser half didn't.
• The repo's own docs assert the behavior this restores: flowx-discover/SKILL.md documents the activities count as "including nested ForEach/If/Switch children", and flowx-convert/SKILL.md documents nested gaps inside a Switch as reported. The code contradicted its own spec.
• The commit that consciously improved control-flow parsing added "Switch" to the container type-sets but never added case-body parsing — an oversight in that change, not an exclusion. No TODO/FIXME anywhere flags Switch as intentionally unsupported.

Fix

• Add switch_cases (case value → activities) and switch_default_activities to AdfActivity, plus a switch_child_activities() helper for a flat view.
• Parse both in parse_activity.
• Descend into them in _classify_activities (inventory) and _walk_activities (lineage).

The translate/bundle path is untouched — it reads raw typeProperties, not these new fields.

Verification

• Verified on the real app0001 export (327 pipelines): activity_count 1,804 → 2,159 (+355 across 38 pipelines); ExecutePipeline calls reachable by the lineage walker 599 → 808 (~209 edges recovered, ~26% of the true control graph — including orchestrator_main's entire fan-out, previously a near-leaf).
• Full suite: 852 passing (+4 new regression tests covering parse, the helper, inventory count, and lineage reachability); ruff + mypy clean.
• 6 pre-existing convert/SetVariable golden-output failures are present on main unchanged (verified identical at the base commit) — unrelated to this change.

Note on stacked branches

This targets main. The in-flight feat/discover-lineage (PR the discover-lineage PR) and feat/discover-insights (PR the discover-insights PR) both sit on the buggy parser, so their lineage/inventory numbers improve once this lands and they rebase.

This pull request and its description were written by Isaac.