[Bug]: --update merge block overwrites .graphify_extract.json, so re-running it or Step 9 stamps a doc whose chunk failed #4811

"""After an --update in which a changed doc's chunk failed, a re-run of the merge block, or Step 9,
stamps the doc as done. Run from the root of a graphify checkout:

    python merge_rerun_repro.py rerun   # run the update merge block a second time
    python merge_rerun_repro.py step9   # run core Step 9 after the update

Renders the Claude skill with tools.skillgen and runs its real blocks through bash.
"""

import json
import os
import re
import subprocess
import sys
import tempfile
from pathlib import Path

REPO = Path.cwd()
sys.path.insert(0, str(REPO))
from tools.skillgen import gen  # noqa: E402

rendered = {a.path: a.content for a in gen.render_all(gen.load_platforms(), only="claude")}
update = re.findall(r"```bash\n(.*?)```", rendered["graphify/skills/claude/references/update.md"], re.S)
core = re.findall(r"```bash\n(.*?)```", rendered["graphify/skill.md"], re.S)
pick = lambda blocks, anchor: next(b for b in blocks if anchor in b)  # noqa: E731
detect = pick(update, "result = detect_incremental(")
populate = pick(update, "'all_files': r.get('files'")
merge = pick(update, "build_merge(")
step9 = pick(core, "Update cumulative cost tracker")

env = {**os.environ, "PYTHONPATH": str(REPO), "GRAPHIFY_OUT": "graphify-out"}
proj = Path(tempfile.mkdtemp())
out = proj / "graphify-out"
out.mkdir()
(out / ".graphify_python").write_text(sys.executable)


def py(source: str) -> str:
    return subprocess.run([sys.executable, "-c", source], cwd=proj, env=env, check=True,
                          capture_output=True, text=True).stdout


def bash(block: str) -> None:
    block = block.replace("INPUT_PATH", ".").replace("IS_DIRECTED", "False")
    subprocess.run(["bash", "-c", block], cwd=proj, env=env, check=True, capture_output=True)


def queued() -> list[str]:
    return json.loads(py("import json\nfrom pathlib import Path\nfrom graphify.detect import detect_incremental\n"
                         "print(json.dumps(sorted(Path(f).name for fl in detect_incremental(Path('.'))"
                         "['new_files'].values() for f in fl)))"))


# A built project: notes.md extracted and stamped, its node in graph.json.
(proj / "notes.md").write_text("# Notes\nWidget\n")
py("from pathlib import Path\nfrom graphify.cli import _stamped_manifest_files\n"
   "from graphify.detect import detect, save_manifest\n"
   "x = {'nodes': [{'id': 'notes_widget', 'source_file': 'notes.md'}], 'edges': [], 'hyperedges': []}\n"
   "save_manifest(_stamped_manifest_files(detect(Path('.'))['files'], x, Path('.')), root='.')")
(out / "graph.json").write_text(json.dumps({"nodes": [{"id": "notes_widget", "label": "Widget",
                                                       "source_file": "notes.md"}], "edges": [], "hyperedges": []}))

# --update: notes.md changed and its chunk failed, so this run's extraction has nothing for it.
(proj / "notes.md").write_text("# Notes\nWidget and Sprocket\n")
bash(detect)
bash(populate)
(out / ".graphify_extract.json").write_text(json.dumps({"nodes": [], "edges": [], "hyperedges": [],
                                                        "input_tokens": 0, "output_tokens": 0}))
bash(merge)
print("after the update's merge block, next --update queues:", queued())

bash(merge if sys.argv[1] == "rerun" else step9)
print(f"after {'a second merge' if sys.argv[1] == 'rerun' else 'Step 9'}, next --update queues:", queued())