Will close dask/dask#10760

TL/DR; dask-expr needs an extra check for a lambda function in _normalize_lambda since free functions are also instances of types.LambdaType (it's actually an alias to types.FunctionType...because, why not?). And failing on Python 3.12 because that's the only one that also includes a dask-expr dependency.

git bisect'd to #10676 (the last tests in that PR unsurprisingly also show the failing 3.12 test)

Running the failing test by itself (test_use_cloudpickle_to_tokenize_functions_in__main__) will work fine, but running the module (python -m pytest -k test_base) will bring the issue out. Then commenting out the two tests added in that PR will also allow everything to pass.

Specifically, running by itself returns the expected bytes but running the module the normalize_token will instead return a string, like '<function inc at 0x7f5e97dad9e0>', giving the error we saw in the failing test.

TypeError: 'in ' requires string as left operand, not bytes

This was because dask-expr's _utils._normalize_lambda registered on types.LambdaType. Adding an extra check for the 'lambda' function name as well as <locals> to avoid cloudpipe/cloudpickle#385 resolves the issue. Just so happens the tests added in the referenced PR managed to import dask-expr and thus added _normalize_lambda to the mix.