Only import runner from edge_impulse_linux

[AIWintermuteAI]

As a side effect of how package init files is done now, a missing PortAudio/OpenCV crashes the whole package import, even for user who did not need audio/image.

With this change users who only need ImpulseRunner can now import edge_impulse_linux without having pyaudio or opencv-python installed. This removes awkward comments like this

# Edge Impulse Linux SDK + pyaudio. The SDK imports `pyaudio` at load time
# (even for image models), and `portaudio19-dev` above is needed to build it.

NB: edge_impulse_linux.audio and edge_impulse_linux.image are no longer auto-loaded as attributes of the package. Code that relied on import edge_impulse_linux; edge_impulse_linux.audio.AudioImpulseRunner would break. However, all the examples use explicit submodule imports (from edge_impulse_linux.audio import ..., from edge_impulse_linux.image import ...), which still works, as Python imports the submodule on demand. So the examples are unaffected.

[AIWintermuteAI]

Ideally we also want to re-write the install to be modular, e.g.

pip install edge_impulse_linux[all]
pip install edge_impulse_linux[image]
pip install edge_impulse_linux[audio]
pip install edge_impulse_linux[minimal]