fix: accept 'args' alias and null value in validate_tool_request

[GratefulDave]

Problem

validate_tool_request raises a ValueError intermittently when the LLM emits "args" instead of "tool_args" in its JSON response:

ValueError: Tool request must have a tool_args (type dictionary) field

The bug is a mismatch between validation and processing:

• process_tools already handles the "args" alias: .get("tool_args", .get("args", {}))
• validate_tool_request only accepted "tool_args" — rejecting "args" before process_tools could normalize it

Also, a null / missing tool_args value raised erroneously, even though process_tools already defaults it to {}.

Fix

# Before
if "tool_args" not in tool_request or not isinstance(tool_request.get("tool_args"), dict):
    raise ValueError("Tool request must have a tool_args (type dictionary) field")

# After
tool_args_value = tool_request.get("tool_args", tool_request.get("args"))
if tool_args_value is not None and not isinstance(tool_args_value, dict):
    raise ValueError("Tool request must have a tool_args (type dictionary) field")

• Accepts "args" as an alias for "tool_args" (matches what process_tools already does)
• Tolerates null / missing value (downstream already defaults to {})
• Still raises if the value is present but not a dict (genuine malformat)

Impact

• Eliminates intermittent ValueError crashes when models use non-standard field names
• No behaviour change for well-formed requests
• 2-line change, no new dependencies