Adds the waiting loan's diff cases, one per class

[johnnyt]

Summary

Adds diff and position cases for a library loan that waits in awaiting_return with its due-date timer handed to the host, while its chart is edited so that the check_in state routing after the wait gains a damaged outcome (a damaged key, an in_repair final state and the transition to it). One case per class, each with expect_diff and expect_compatible_at for the position the loan waits at:

Case	Edit to awaiting_return	Class	Predicate
loan_waiting_across_a_damaged_arm	none: unchanged, id and content alike	compatible	true
loan_waiting_across_a_mapped_rename	renamed to awaiting_check_in, with a mapping	mapped	false (it takes no mapping)
loan_waiting_across_a_regrouping	id kept, gathered with overdue under a new compound state	breaking	false (parent differs)

The expected reasons are the ones Statifier.Chart.diff/3 answers today, order included; test/corpus/diff_cases_test.exs compares them over the committed corpus.

• Corpus re-emitted with mix statifier.corpus: manifest.json's statifier case_count goes from 28 to 31, the corpus hash moves; mix statifier.corpus --check passes.
• conformance/README.md: the diff-case paragraph now says the pairs take a hold_queue or a loan chart, and a new paragraph describes the loan pairs.
• No lib/ change (git diff origin/main --stat -- lib is empty). No changelog fragment: changelog.d/README.md excludes conformance fixtures. No release.

Provenance

• In the Compatible case awaiting_return is unchanged, id and content alike: the damaged outcome is the only edit, which is what makes the pair compatible. An id change is the Mapped case.
• The Breaking case moves the waiting state under a new parent rather than removing it, so it differs from the existing hold_queue rename-without-mapping case.
• The three cases are also ratcheted into test/passing_tests.json with mix test.baseline add, and registry.json is re-emitted, so the registry's statifier claim names them as it names every other authored case. The file map did not name this; it keeps the claim complete.
• The execution waits at its initial state (steps is empty), as hold_queue_identical and hold_queue_data_added do; the library world gains no event.

Checks

• mix quality, full, green on the head tree (re-run after the description correction): 3,067 of 3,067 tests passed, Dialyzer, Credo, conformance corpus and regression ratchet all passed. mix quality --profile merge green on the first commit; the ADR judge skipped (no file in a judged scope).
• Sabotage, each restored from a copy and re-run green: the compatible case's expect_compatible_at flipped to false in the corpus file -> the position test red, naming that case; the breaking case's class changed to compatible -> the diff test red, naming that case.

Review tier: contract (a corpus expectation).