Compare two sets as a library function and a command

[peterbjohnson]

tests/test_against_convert.py holds the comparison of two sets - question by question, part by part, main text, part text and worked solution, with whitespace runs folded, image references compared by file name and a lone empty part dropped - and the differs.txt convention that records known differences per folder. The agent needs the same comparison to check a built set against the Lambda Feedback export it should reproduce, so it belongs in the package.

Move it into in2lambda.compare: differences(built: Set, expected: Set) -> list[str] returning the lines the test now builds (Question 1 "", part (a), worked solution: the draft says ... and convert says ..., with the two sides named by arguments left_name/right_name defaulting to "the draft" and "convert"), and known(path) -> list[str] reading a differs.txt with # ticket comments. Keep the normalisation rules as they are and document them in the module docstring; tests/test_against_convert.py imports both and loses its private copies. Add in2lambda compare BUILT_ZIP EXPORT_DIR [--known differs.txt] to the CLI, printing each difference and exiting 1 where the differences are not exactly the known ones, so a person can compare a set with an export by hand. A unit test in tests/test_compare.py covers the three normalisations and the known-differences check with two small sets built in the test. Done when the against-convert suite passes unchanged in outcome and the CLI compares the ME2 export fixture with itself as identical.

---

Workbench ticket t55.