fix(frontierchallenge): make open-track selection reproducible

[fzp0424]

Summary

• resolve one registry-validated effective task set and reuse it for staging, ORCA preflight, verifier handling, resume, and Harbor
• keep stale or excluded task directories out of evaluation and classify task_098 as open from metadata rather than instruction text
• dereference Hugging Face cache symlinks while staging
• pin solve and reference dataset revisions independently
• clarify the execution-based ORCA track boundary across the runtime documentation

Hugging Face release

• solve: 8a73bb02a5709aa007e1a146d0cdc5d36ef5ea4d
• reference: 59fbf007ad5c6816caf7d379eef32d9fb384edea
• removed the obsolete solve-side statement unsealer; instructions remain plaintext

Verification

• 45 FrontierChallenge tests pass
• fresh-cache verification passes for 97 solve tasks and 97 encrypted verifiers
• task_098 runs through the runner without ORCA preflight, even with a stale licensed task in staging
• Harbor receives only the exact selected task and staged HF files contain no broken symlinks
• public leak, restricted-software, shell syntax, Python compile, and diff gates pass

Addresses #27 and #28.

[zhanghanduo]

Small upgrade-path edge case: the stage cache identity is unchanged from the previous runner, so an existing default stage created with cp -a can still pass the reuse check and skip the new cp -aL. If that cached copy contains Hugging Face cache-relative symlinks, they may remain broken after upgrading unless the evaluator uses --force-restage or a fresh stage directory.

It may be worth adding a staging-format/version value to .frontier-source, checking cached tasks for symlinks before reuse, or otherwise forcing a one-time restage.