Fix remaining crawl issues and support native Windows workflows

[PavloSEO]

Native SQLite/project workflows now run with Windows OS-backed writer locks, portable snapshots and file publication, and a python -m seohead entry point. Windows CI exercises Unicode/spaced paths, writer contention, native retry/reanalysis, CLI contracts, and real stdio MCP; POSIX locking and directory flushing remain intact.

The same delivery fixes the open operational issues:

• Explicit, off-by-default retention of no-store evidence with recorded acknowledgement; credentialed data remains excluded.
• Fresh anonymous-session resume, including compatible old cookie-latched scans, without saving cookie values or relaxing configured-credential guards.
• Consistent --scan paths and --input JSON, with warned backward-compatible aliases.
• Root sitemap discovery through robots and the conventional fallback; all-failed parsing now fails instead of returning a clean success.
• Current primary Google citations for structured data, robots and crawler guidance, including the May 2026 FAQ retirement.

Validation:

• Local regression/interface/documentation gates passed; Ruff and the covered mypy surface passed.
• Real Windows runner: OS locks, Unicode/spaced project paths, native scan v1/v2 snapshot/retry/reanalysis, CLI registration and real stdio MCP (170 tests passed).
• Isolated Linux VPS, Python 3.14.4: 99 targeted tests passed. The installed wheel also passed a real loopback HTTP crawl with anonymous session cookies and acknowledged no-store retention, snapshot, requeue/resume and source-preserving offline reanalysis.
• Exact-head full CI passed on Python 3.10/3.12/3.13; Windows, contracts, package, Docker and CodeQL also passed. No paid provider calls or changes to running VPS services.

Closes #700.
Closes #701.
Closes #702.
Closes #703.
Closes #704.
Closes #705.