fix: fail fast on provider-wide X outage

[hiidkaboutyou-spec]

Production incident

Current main Daily runs repeatedly hit Cloudflare HTTP 403 at x.com before GraphQL, then spend the 15-minute monitor budget retrying each configured source. Nightly Fanfic remains healthy.

Change

• publish the optional X preflight result to later Actions steps through GITHUB_ENV;
• when that immediately preceding probe is offline, skip the doomed provider scan;
• keep every enabled source explicitly failed (provider_preflight_offline);
• retain the collection cursor and existing observations;
• continue Telegram polling, pending delivery, state checkpointing, and production-outcome emission;
• document the live incident and the failed curl/browser-TLS experiment.

This does not claim X collection is restored and cannot create COMPLETE evidence.

Validation

• python -m pip check
• python -m compileall -q app tests tools
• python -m app --check
• full Python suite: 1,161 passed, 4 skipped
• Rust tooling was unavailable locally; repository CI remains the Rust gate.

Next operational gate

Probe anonymous X egress on a non-GitHub host without credentials. Only migrate collection after the host returns valid X HTML/GraphQL evidence; never move cookies into an unproven probe.