refactor: trace_impact.py breaches the file and function size caps, with zero test coverage

[cdeust]

Surfaced while fixing the py/constant-conditional-expression alert in this file (#46). Recorded per §14.3 as a deferral with acceptance criteria, because a behaviour-preserving decomposition of a 329-line async orchestrator belongs in its own PR (§15.1: "never mix the enabling refactor and the feature in one PR"), not inside a CodeQL-triage change.

Current state (measured on fix/46-codeql-tail, after the if True: removal)

cortex_viz/server/trace_impact.py is 559 lines — over the 500-line file cap (§4.1).

Four functions breach the 50-line function cap (§4.2):

Function	Lines	Limit
_impact_for_graph()	329	50
_run() (nested inside it)	300	50
_ast_and_impact()	74	50
_to_repo_relative()	58	50

Reproduce:

python3 - <<'PY'
import ast, pathlib
src = pathlib.Path("cortex_viz/server/trace_impact.py").read_text()
for n in ast.walk(ast.parse(src)):
    if isinstance(n, (ast.FunctionDef, ast.AsyncFunctionDef)):
        span = n.end_lineno - n.lineno + 1
        if span > 50:
            print(f"{n.name}() = {span} lines")
PY

Why this is more than a style breach

No test references this module. grep -rln "trace_impact\|_impact_for_graph" tests/ returns nothing across 239 test files. The 300-line _run() issues ~12 distinct Cypher queries and a capped get_context fan-out against a live AP bridge; every one of its failure arms (absent file node, empty members, bridge error, the N>20 fallback path) is unexercised. That is the §13.1 A3/F1 gap that makes the decomposition worth doing rather than cosmetic — the size cap is the symptom, the untested query orchestration is the defect.

The if True: that wrapped the whole body (removed in the #46 PR) is evidence the block was never revisited after whatever condition it once carried was dropped.

Acceptance criteria

• trace_impact.py is under 500 lines, or split along a concern boundary into modules that each are
• No function in the resulting modules exceeds 50 lines
• The Cypher query set is separated from the result-shaping so each is testable without a live bridge
• Every failure arm of the impact path has a test asserting its observable effect: file absent from graph → None; empty members; bridge raising; the >N member fallback to the per-file Cypher path
• The refactor is behaviour-preserving and ships as its own PR — the existing suite passing unchanged is the proof, with the new tests added on top
• git diff -w reviewed to confirm no accidental semantic change in moved blocks