Accept `np.ndarray` as a typehint for actions/properties

We use numpy a lot, so it would be great to be able to type hint numpy properly.

Currently we have to use: from labthings_fastapi.types.numpy import NDArray which works for sending things over HTTP but means the type hints are wrong when using the action from within the server.

Proposal

• Allow np.ndarray
• On seeing np.ndarray LabThings should convert it to a pydantic model this model can have the necessary data to reconstruct the array including dimensions and types

class NDArrayModel(BaseModel):

    shape = list[int]
    d_type = Literal["bool"], Literal["int"], Literal["float"], Literal["uint8"], Literal["uint16"]
    data = list[int]|list[bool]|list[float]

This way we can serialise the data as a 1D list along with the information for how to contruct the correct array. In a ThingClient the array should be able to be reconstructed. In Javascript or any other language there is enough information to process the result.

Start with the existing labthings_fastapi.types.numpy.NDArray type and the ThingClient path that reconstructs returned values. Trace how action and property type hints are converted for HTTP, then define what metadata is needed for shape, dtype, and flattened data. Done means np.ndarray is accepted and arrays can be serialized and reconstructed across the server and client.