fix: remove stale type key from union schemas

[thechandru]

Problem: When using union types with parameterized containers (e.g. list[str] | None), the generated JSON schema includes a conflicting type: object key alongside anyOf, causing LLMs to return stringified arrays instead of proper JSON arrays.

Reproduction:

from toolslm.funccall import get_schema

def f(items: list[str] | None = None):
    "Docstring"
    pass

get_schema(f)['input_schema']['properties']['items']
# {'type': 'object', 'description': '', 'default': None,
#  'anyOf': [{'type': 'array', 'items': {'type': 'string'}}, {'type': 'null'}]}
# 'type': 'object' conflicts with anyOf per JSON Schema spec

Fix: In _process_property, pop the type key when anyOf is present. Per JSON Schema spec, keywords at the same level are ANDed — so type: object alongside anyOf creates a contradictory schema.

After fix:

{'description': '',
 'default': None,
 'anyOf': [{'type': 'array', 'items': {'type': 'string'}}, {'type': 'null'}]}

Verified: Tested end-to-end with Lisette — Claude now returns proper JSON arrays instead of stringified arrays. All nbdev-test tests pass.

[thechandru]

@jph00 - this is my second attempt at #74

[jph00]

Thanks @thechandru for the revised PR. It looks like you accidentally deleted all the outputs. Can you put them back? Try to keep your diff nice and clean! :D

[thechandru]

@jph00 - I am not entirely sure how I managed to do that.

I did a run-all and then did a commit from solveit and that seems to have restored outputs. Not sure if I got it back to a sane diff for review. Happy to RTFM - not sure where to look :-)

[jph00]

Very good! :)