Extend the pinned-release self-consistency test to release_due's new-machine discovery #168

Follow-up from #163 (Codex review of 0879839, finding "Apply recursive discovery to pinned releases").

tests/test_release_pin_contents.py::test_pinned_release_is_self_consistent still uses release_due.commit_paths(), which scans only the root new-host documents with the extension-based PATH_RE. #163's drift discovery (new_machine_files()) also follows relative Markdown links, linked directories, bare repository paths, and install guides (Markdown under adoption/ or recipes/) to closure — but only against the working tree, where every candidate can be filtered by existence.

At a commit, a missing file cannot be told from prose by existence. A trial in #163 (reverted) showed two false-positive classes:

• guide prose naming another repository's file (recipes/claude-native-ultracode.md: "agent-lab's docs/native-token-workflow.md");
• JSON method names in fenced blocks (tools/call, tools/list) that start at a real top-level directory.

Following only the guides' relative links at the commit was precise (v2026.09.23.1 and main: 0 missing) but misses a guide that names an extensionless runner in prose.

Needed: a reference form that is precise at a commit — for example, an explicit machine-readable list of each guide's install inputs, or a marker for runnable paths — then use it in the self-consistency test.

🤖 Generated with Claude Code