[BUG] A damaged run record crashes the resume listing for healthy runs

Describe the bug

One damaged or incompatible run can prevent Strix from listing all prior runs, including healthy ones. strix -n --resume raises an uncaught UnicodeDecodeError or TypeError instead of showing the normal run-name help. The interactive resume picker reads the same summaries.

There are two gaps in the existing best-effort reader:

• Invalid UTF-8 in either run.json or vulnerabilities.json bypasses _load_json()'s JSON/I/O fallback.
• A JSON object such as {"targets_info": 123} parses successfully, but _describe_target() tries to iterate the number.

Malformed JSON and non-object run records already receive empty/default summaries. These cases should follow that same policy, while healthy sibling runs stay visible.

To Reproduce

On current main 09536cf1481731c790995bd8060db19a3b75ff48, with frozen Python dependencies, this setup-only probe creates temporary records and calls the same listing used by the CLI. No credentials, LLM, Docker or scan target are used:

LITELLM_MODE=PRODUCTION LITELLM_LOCAL_MODEL_COST_MAP=True STRIX_TELEMETRY=0 uv run --frozen python - <<'PY'
import json
import os
import tempfile
from pathlib import Path
from strix.interface.cli_args import resume_run_list_message

with tempfile.TemporaryDirectory() as directory:
    os.chdir(directory)
    healthy = Path("strix_runs/healthy_1111")
    damaged = Path("strix_runs/damaged_2222")
    healthy.mkdir(parents=True)
    damaged.mkdir()
    (healthy / "run.json").write_text(json.dumps({"status": "completed"}), encoding="utf-8")
    for payload in (b"\xff", b'{"targets_info": 123}'):
        (damaged / "run.json").write_bytes(payload)
        try:
            print(resume_run_list_message("Prior runs"))
        except (UnicodeDecodeError, TypeError) as error:
            print(type(error).__name__, str(error))
PY

Actual: UnicodeDecodeError for the first record and TypeError: 'int' object is not iterable for the second. Neither listing reaches the healthy run.

Expected behavior

List healthy runs normally, retaining the existing unknown/default summary behavior for unreadable metadata. When targets_info is not an array, use the existing workspace/instruction fallback. This should not rewrite artifacts or claim that a damaged run itself can be resumed successfully.

System Information:

• OS: macOS, arm64
• Strix: current main 09536cf1481731c790995bd8060db19a3b75ff48, package 1.7.0
• Python: 3.12.14; frozen uv.lock
• LLM: none; offline fixtures only

Additional context

I checked the all-state PR/issue inventory, including the existing resume work #1421, #1450 and #1621. #1621 concerns restoring the API-spec mount after a run is selected; this failure occurs earlier, while enumerating runs. A small reader-validation fix with CLI and summary regressions is in preparation.