Export expanded-case factor metadata in hardness runs

[GISMO-1]

Motivation

• Attach structural tags (residue, SPF/cofactor, semiprime kind) to expanded cases at source so large-range hardness runs can collect classification data without separate profiling scans.
• The SPF search must be bounded to keep per-case work predictable and to scale to huge ranges.
• Keep factoring/primality logic reusable and deterministic and avoid changing existing expanded export behavior when the new option is not supplied.

Description

• Add a new CLI option --export-expanded-meta PATH and a bound option --expanded-factor-bound INT (default 20000) to the strausforge hardness command and wire them through to run_hardness in hardness_core.py.
• Introduce a new reusable module src/strausforge/factor_meta.py with smallest_prime_factor_bounded, a trial-division primality helper, semiprime_kind_from_spf, and write_jsonl_record to keep factoring helpers separate from CLI logic.
• Extend run_hardness to optionally emit a second JSONL stream (one object per expanded case, deterministic ordering by increasing n) containing the fields n, res48, identity, t_used, window_used, proc_heuristic, spf, cofactor, and semiprime_kind, with SPF search capped by min(expanded_factor_bound, isqrt(n)) and spf == 0 when no factor is found within the bound.
• Update README.md and CLI examples to document --export-expanded-meta and --expanded-factor-bound and note the intended research purpose.
• Add unit tests in tests/test_factor_meta.py covering smallest_prime_factor_bounded, semiprime classification (p*q and p^2), the JSONL writer producing one object per line, and that the expanded-meta JSONL is emitted in sorted/evaluation order and matches the expected schema.

Testing

• Installed dev dependencies via python -m pip install -e '.[dev]' which completed successfully.
• Ran targeted tests with pytest -q tests/test_factor_meta.py tests/test_hardness_core.py tests/test_cli.py which passed (targeted run completed successfully).
• Ran the full test suite with pytest -q and observed all tests passing (68 passed, 1 skipped).

---

Codex Task