Normalize unit_type field shape across parsers so EntryPointDetector's unit-type check works for JS/Go

[ar7casper]

Context

utilities/agentic_enhancer/entry_point_detector.py:176 reads:

unit_type = func_data.get('unit_type', '')
if unit_type in ENTRY_POINT_TYPES:
    reasons.append(f'unit_type:{unit_type}')

This is "Check 1" of entry-point detection. It only fires when the function's metadata has a unit_type key in snake_case.

The gap

Different parsers emit different shapes:

Parser	unit_type key shape
Python (function_extractor.py)	unit_type (snake_case) ✅
C / Ruby / PHP (builder.export())	unit_type (snake_case) ✅
Zig	unit_type (snake_case) ✅
JavaScript (typescript_analyzer.js)	unitType (camelCase) ❌
Go (normalized in parsers/go/test_pipeline.py)	unitType (camelCase) ❌

For JS and Go scans, func_data.get('unit_type', '') returns '' and Check 1 silently fails. Only Check 3 (input patterns like req.body / req.params) actually catches anything.

Practical impact

• Express handler bodies that touch req.body get caught by Check 3 — fine in practice
• Minimal Express middleware (async (req, res, next) => next();) has no body content matching input patterns → silently dropped from analysis
• route_middleware units (added in fix: extract Express.js anonymous route handler callbacks #49) have the same issue — Check 1 should catch them but doesn't due to camelCase

#49's regression test test_route_middleware_is_entry_point passes in isolation (constructs unit_type snake_case directly) but the full JS pipeline gap remains.

This is documented

_docs/internal/parser-issues.md items #23-26 describe the broader normalization gap. _docs/internal/how-openant-works.md Room-for-Improvement #10 lists "Fix test_pipeline.py normalization in ALL parsers" as the highest-impact fix.

Proposed approaches

Option A — Normalize at parser output time (preferred)

Each parser's test_pipeline.py should emit unit_type (snake_case). Python / C / Ruby / PHP / Zig already do. JS and Go don't.

Concrete fix:

• parsers/javascript/test_pipeline.py:497: normalize unitType → unit_type before passing to EntryPointDetector
• parsers/go/test_pipeline.py:296: change 'unitType': ... → 'unit_type': ...
• Same for the call_graph.json writes added in feat: LLM review stage for enhanced reachability detection #50 (so apply_reachability_filter re-filter also gets snake_case)

Option B — Make EntryPointDetector accept both

unit_type = func_data.get('unit_type') or func_data.get('unitType', '')

One-line change but doesn't align the underlying schema.

Recommendation

Option A. Establish snake_case as the canonical schema, update tests/test_call_graph_output.py to assert unit_type exists in each function entry.

References

• utilities/agentic_enhancer/entry_point_detector.py:176
• parsers/javascript/test_pipeline.py:497
• parsers/go/test_pipeline.py:296