Fix: Handle empty schema type in MCP tool manifests

[ganeshyb]

Problem

Some MCP tool servers (e.g. HuggingFace MCP Server) return tool manifests where certain parameters have an empty string for their JSON Schema type field (e.g. "seed": {"type": ""}). This causes a pydantic_core.ValidationError when the SDK deserialises the ListFoundryConnectedToolsResponse, because SchemaType only accepts the six standard JSON Schema types (string, number, integer, boolean, array, object).

Error

pydantic_core._pydantic_core.ValidationError: 1 validation error for ListFoundryConnectedToolsResponse
ResultType.tools.0.manifest.9.parameters.properties.seed.type
  Input should be 'string', 'number', 'integer', 'boolean', 'array' or 'object'

Fix

1. azure-ai-agentserver-core - _models.py

• Changed SchemaProperty.type from SchemaType to Optional[SchemaType] with a default of None
• Added a model_validator(mode='before') that coerces empty type strings ("") to None

2. azure-ai-agentserver-agentframework - _foundry_tools.py

• Added a guard in FoundryToolClient._create_ai_function that skips parameters with type is None and logs a warning, instead of crashing with AttributeError: 'NoneType' object has no attribute 'py_type'

Testing

• Verified locally with HuggingFace MCP Server (previously failing) — now works
• Verified MicrosoftLearn MCP Server still works (was never affected)
• No existing tests broken