fix: bring workflow_graph_ast_* infrastructure under the 40-line/method cap

[cdeust]

Summary

Issue #275 (the 300-line/file cap on workflow_graph_source_ast.py) was
already resolved on main by PR #278's follow-up commit (0702b77) — an
independent split into ap_sync_loop.py / workflow_graph_ast_symbols.py
/ workflow_graph_ast_edges.py / workflow_graph_ast_response.py landed
and closed #275 while this branch was still based on the pre-#278 tree.
Rebasing this branch's own (differently-named) split onto that state would
just duplicate already-merged work and conflict with it — docs/module- inventory.md already lists all four modules — so this PR now keeps
main's file layout and instead fixes what was actually wrong: a false
ledger row in this PR's prior revision claimed "§4.2 pass" while AST
measurement found 4 functions over the 40-line/method cap (up to 53
lines), and once the rebase touched these files, AST measurement found 3
more pre-existing violations in the already-merged #278 split (up to 193
lines) — boy-scout material in the same touched files (coding-standards.md
§14).

Pure extract-method, behavior-preserving, no file renames/moves:

File	Split
ap_sync_loop.py	_drain_pending_tasks (90) → _loop_is_drainable (guard) + _run_task_drain/_cancel_and_await_pending (schedule/await)
workflow_graph_ast_response.py	as_list (46) → _zip_columns_rows + _from_legacy_shape; added build_path_tails (was duplicated inline in both symbols.py and edges.py — extracted once)
workflow_graph_ast_symbols.py	_symbol_type_from_label (41) → a dispatch dict (_LABEL_TO_SYMBOL_TYPE, OCP-friendly over an if-chain); symbol_batches_async (127) → _where_for_tails/_build_symbol_query/_parse_symbol_batch; verify_symbols_async (44) → _collect_all_symbol_names/_qualname_matches
workflow_graph_ast_edges.py	edge_batches_async (193) → rel-table enumeration split by family (_calls_and_member_rel_tables/_import_rel_tables/_uses_rel_tables) + _run_edge; _run_edge (94) → _build_edge_query/_edge_row_to_dict; _edge_row_to_dict (46) → _split_edge_endpoints/_parse_provenance

Public import paths are unchanged (WorkflowGraphASTSource, _SyncLoop,
_SYMBOL_LABELS all still resolve from workflow_graph_source_ast, which
this PR does not touch — it had no over-cap functions).

Completion Ledger (coding-standards.md §13.2)

Item	Evidence
§4.1 file ≤300 lines, all 5 infra files	AST measurement (command below): every file ≤300 lines — command output quoted in "Verification"
§4.2 method ≤40 lines, all functions in all 5 infra files	AST measurement: every function ≤40 lines (the claim this PR was refused for is now TRUE by measurement, not restated)
_drain_pending_tasks behavior unchanged	tests_py/infrastructure/test_workflow_graph_source_ast.py::TestDrainPendingTasks (5 tests) — all pass unmodified
_SyncLoop construction/close/wedge-timeout contract unchanged	TestSyncLoopConstruction, TestSyncLoopCloseRealLoop, TestBoundedWaitTimeout, TestSingleReaderOwnership — all pass unmodified
as_list/build_path_tails normalization unchanged	TestIncrementalYield, TestNoFullMaterialization (exercise as_list indirectly via iter_symbols/iter_ast_edges) — pass unmodified
_symbol_type_from_label dispatch-dict equivalence	Verified interactively (function→function, struct→class, Import→import) — same 3 branches as the original if-chain, now table-driven
symbol_batches_async/verify_symbols_async per-graph skip-on-error unchanged	tests_py/infrastructure/test_ble001_sweep_infrastructure.py::TestWorkflowAstGraphSkips — pass unmodified
edge_batches_async/_run_edge unchanged	TestIncrementalYield::test_edges_yield_incrementally — pass unmodified
No regression from the refactor	Paired mutmut comparison, same 3 test files, main's pre-refactor originals vs. this PR's files — see "Mutation testing" below

Verification

$ python3 - <<'PY'
import ast, pathlib
files = ["mcp_server/infrastructure/ap_sync_loop.py",
         "mcp_server/infrastructure/workflow_graph_ast_edges.py",
         "mcp_server/infrastructure/workflow_graph_ast_symbols.py",
         "mcp_server/infrastructure/workflow_graph_ast_response.py",
         "mcp_server/infrastructure/workflow_graph_source_ast.py"]
for f in files:
    src = pathlib.Path(f).read_text(); n = len(src.splitlines())
    assert n <= 300, f"{f}: {n} lines"
    for node in ast.walk(ast.parse(src)):
        if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef)):
            ln = (node.end_lineno or node.lineno) - node.lineno + 1
            assert ln <= 40, f"{f}::{node.name}: {ln} lines"
print("ALL CLEAR")
PY
ALL CLEAR

• ruff check + ruff format --check: clean on all touched files.
• pyright: 0 errors/warnings on all 5 infra files + the two consuming
  handlers (unified_search.py, wiki_verify.py).
• pytest full suite: 6751 passed, 5 skipped, 0 modified/added (pure
  refactor — same behavior, same tests).
• scripts/check_doc_claims.py / scripts/generate_repo_badges.py --check: both green, untouched (issue fix: eliminate the six-file test-count conflict class with a monotone floor (issue #293) #294's monotone floor is
  authoritative on main; this PR does not edit any doc/badge file).

Mutation testing (§12) — paired before/after comparison

Scoped mutmut run (the 5 infra files vs. test_workflow_graph_source_ast.py

• test_ble001_sweep_infrastructure.py + test_s110_sweep_infrastructure.py),
  compared against an identical scoped run against main's current (already
  fix: drain pending tasks before stopping the AP sync loop (issue #258) #278-split, pre-this-PR) files, same test files:

	main (pre-refactor)	This PR (post-refactor)
Total mutants	876	776
Killed	244	299
Survived	460	272
No-tests-in-scope	168	199
Timeout	4	6

The refactor reduces survivors (460→272) and increases kills
(244→299) — consistent with deduplicating the path-tails logic
(build_path_tails, previously copy-pasted in both symbols.py and
edges.py) and replacing _symbol_type_from_label's if-chain with a
dispatch dict (fewer independently-mutable branches). No regression.

The remaining ~272 survivors are a pre-existing, unchanged-in-kind
gap: this suite (by its own docstring) targets the streaming/ordering/
bounded-wait contract, not Cypher query-string construction or
edge/symbol row-parsing detail — identically true before and after this
PR (main's pre-refactor score already showed the same 460-survivor
shape). Closing it is a dedicated test-hardening effort (~300 new
assertions across query construction and row parsing), out of scope for
a behavior-preserving cap-compliance fix (coding-standards.md §15.1: a
large prerequisite is sequenced as its own PR, not mixed into this one).

Test plan

• pytest tests_py/infrastructure/test_workflow_graph_source_ast.py tests_py/infrastructure/test_ble001_sweep_infrastructure.py tests_py/infrastructure/test_s110_sweep_infrastructure.py — 64 passed, 0 modified, 0 added
• Full suite: 6751 passed, 5 skipped
• ruff check + ruff format --check clean
• pyright 0 diagnostics
• AST §4.1/§4.2 measurement — all 5 files clean (see "Verification")
• Paired mutation-testing comparison — no regression (see above)
• Live import smoke test: WorkflowGraphASTSource, _SyncLoop,
  _SYMBOL_LABELS, symbol_batches_async, verify_symbols_async,
  edge_batches_async, as_list, build_path_tails all resolve

Co-Authored-By: Claude noreply@anthropic.com