Support Annotated types for field descriptions in schema generation

[boboletus]

toolslm/toolslm/funccall.py

Line 40 in b3b7f8b

def _param(

When using lite_mk_func with dataclasses, field descriptions are not populated from type hints. A fix solve.it help figured out is to support typing.Annotated to specify field descriptions:

from typing import Annotated, Optional

@dataclass
class Adult:
    "How adults are supposed to help in this game"
    adult_role: Annotated[str, "facilitator | demonstrator | player | observer"]
    adults_required: Annotated[int, "Number of adults needed"]
    losing_gracefully: Annotated[Optional[bool], "Can adult elegantly lose? None if not competitive"]

But this would cause it to produces empty descriptions and incorrect types ('type': 'object' instead of 'string').

{'type': 'function',
 'function': {'name': 'Adult',
  'description': 'How do adults supposed to help in this game',
  'parameters': {'type': 'object',
   'properties': {'adult_role': {'type': 'object', 'description': ''},
    'adults_required': {'type': 'object', 'description': ''},
    'losing_gracefully': {'type': 'object', 'description': ''}},
   'title': 'Adult',
   'required': ['adult_role', 'adults_required', 'losing_gracefully']}}}

This requires modifying _param and _process_property in funccall.py.

With solve.it's help I explored the fix a bit here: https://share.solve.it.com/d/75ab5d185e67486d066af43420ea2eba