Fix Windows receipt filenames and add flaky test judge

[liomilet4-png]

Summary

Frantic bounty #66: add a flaky-test-judge public skill and fix the Windows receipt-store path issue that blocked local harness runs.

• Store canonical sha256:<digest> receipts on disk as Windows-safe sha256-<digest>.json files while preserving canonical receipt IDs in receipt JSON.
• Map safe filenames back to canonical receipt IDs for list/index/history paths, and keep non-Windows legacy colon filenames readable for exact reads.
• Update receipt-store and CLI test helpers for safe receipt filenames, including the Windows MCP dogfood fixture command.
• Add the skills/flaky-test-judge package with two harness cases: quarantine_justified and missing_run_history.

Validation

• git diff --cached --check
• cargo test --manifest-path crates\Cargo.toml -p runx-runtime --test integration receipt_store
  • 25 passed
• cargo test --manifest-path crates\Cargo.toml -p runx-runtime --test integration skill_run::
  • 21 passed
• cargo test --manifest-path crates\Cargo.toml -p runx-runtime --test integration journal_history::
  • 14 passed
• cargo test --manifest-path crates\Cargo.toml -p runx-cli --test integration skill::native_skill_pauses_and_resumes_with_run_id
  • 1 passed
• cargo test --manifest-path crates\Cargo.toml -p runx-cli --test integration mcp_dogfood::mcp_native_binary_dogfoods_streaming_skill_calls_and_receipts
  • 1 passed
• npx -y @runxhq/cli@0.6.13 skill inspect .\skills\flaky-test-judge\SKILL.md --json
  • status: ok
• cargo run --manifest-path crates\Cargo.toml -p runx-cli -- harness .\skills\flaky-test-judge --receipt-dir receipts_66_final_verify --json
  • status: passed, case_count: 2, assertion_error_count: 0

Notes

• Receipt IDs remain canonical sha256:<digest> values; only the local filesystem name changes for Windows compatibility.
• Directory fsync is a no-op on Windows because opening a directory as a file fails there; non-Windows durability behavior is unchanged.
• cargo fmt --manifest-path crates\Cargo.toml --all --check was not run successfully because the local Windows toolchain is missing the rustfmt component.