Reuse scaffold identity helpers in pin and gitignore checks

[Deicyde]

Follow-up to #96. scaffold.py defined _cross_interface_identity twice, with the second definition shadowing the first. Its _file_identity also duplicated the existing _node_identity/_is_link helpers. This change drops the duplicates.

Also in this change:

• The two clean-checkout git diff commands in plugin_pin are built from shared parts. The argv is the same as before.
• hashlib constructors are called directly; requires-python is 3.10.
• The hash length comes from _committed_scaffold_entries.

A second commit factors the scaffold copy, commit, and _here/_TEMPLATES patching that the plugin_pin tests repeated into _git_output, _autoform_checkout and _run_from. Each test keeps its own assertions.

A third commit fixes a regression from the first. _git_blob_id called hashlib.new directly, so an interpreter without SHA-1 or SHA-256 raised ValueError out of plugin_pin instead of leaving the pin unset, as main does. It now returns None in that case, and test_plugin_pin_is_empty_when_git_digests_are_unavailable pins it. Without the fix, that test fails.

Two flags in _git_checkout_clean are not pinned by any test on this branch: the --cached half of the check, and --no-replace-objects. Dropping either one still leaves all 22 plugin-pin tests in tests/test_scaffold.py passing. That includes test_plugin_pin_ignores_git_replacement_objects, which the separate blob comparison in _committed_scaffold_entries still satisfies.

The argv matches main's, and the test refactor keeps every assertion, so main probably has the same gap. I did not rerun the mutation on main. This PR does not close the gap.

Tests: ruff check is clean. On dd90820: tests/test_scaffold.py 100 passed. The digest test fails without the third commit.

[Deicyde]

Exact-head review at a38dbb49: helper reuse preserves Git argv, canonical executable-bit handling, descriptor/named identity checks, link refusal, and fail-closed digest behavior. _committed_scaffold_entries already guarantees one object-ID length and unique paths, so the simplified comparison assumptions hold. Test helper extraction preserves assertions; all nine CI checks and the 100-test scaffold suite pass. No blocker remains. Non-blocking follow-up: directly pin the --cached and --no-replace-objects argv invariants, which this PR correctly preserves but existing behavior tests do not uniquely exercise.

[Deicyde]

Post-merge notes on a38dbb4, which merged as f8465bc. autoform_cli/scaffold.py and tests/test_scaffold.py haven't changed on main since, so line numbers are at main 71b251c.

Verdict: nothing to fix. This agrees with the review above. Two optional golf items on private helpers:

1. scaffold.py:399 re-implements the executable-bit rule (0 lines).
   • It computes "100755" if mode & 0o111 else "100644" inline.
   • _canonical_template_mode (:370-373) owns the same rule for the template surface (:380), the running-vs-checkout comparison (:472-473) and generated files (:791).
   • The output is identical today, but if the helper changes, this comparison won't follow it.
   • Fix: else {0o644: "100644", 0o755: "100755"}[_canonical_template_mode(mode)],. That is main's form before this PR (dd90820:442), indexed instead of .get.
2. _git_blob_id can't take its unknown-length branch (2 lines; it predates this PR).
   • Its only caller (:394-400) passes the length of an object ID that _committed_scaffold_entries has already matched against [0-9a-f]{40}|[0-9a-f]{64} (:339).
   • Fix: replace :358-360 with algorithm = {40: "sha1", 64: "sha256"}[length].
   • Keep the hashlib.new(...) / except ValueError fallback, which test_plugin_pin_is_empty_when_git_digests_are_unavailable (tests/test_scaffold.py:953) pins.

Not run locally.

Posted by PR swarm: PR Swarm Lead