test all data types on all canvas backends

[tlambert03]

Just spent a frustrating amount of time debugging what ended up being an issue with passing an unsupported data type to the vispy image node. there was a vague warning in the logs that I overlooked:

/Users/talley/dev/self/ndv/src/ndv/views/_vispy/_array_canvas.py:485: UserWarning: Downcasting integer data to uint16.
  data = _downcast(data)

but the phenotype was that, even though I was passing in new data, the viewer stayed static. Long story short, we need to do comprehensive basic testing to ensure that all possible numpy data types work and/or are cast as needed for the canvas backend

[tlambert03]

ok, the stupid stupid answer was that we downcast data when creating the image texture, but not when updating:

    def set_data(self, data: np.ndarray) -> None:
        if not data.ndim == self._ndim:
            warnings.warn(
                f"Got wrong number of dimensions ({data.ndim}) for vispy "
                f"visual of type {type(self._visual)}.",
                stacklevel=2,
            )
            return
        self._visual.set_data(_downcast(data))  # <------------------------- need downcast here