Switch from nvidia/cuda to plain ubuntu:22.05 base image

Since this is a pure Python environment, the CUDA toolkit isn't really
necessary. The various Python packages that can use CUDA will download
and install the CUDA dependencies they need.  This shaves off at least
8gb from the final image.

The Python package install order needed to be tweaked to ensure onnxruntime,
tensorflow and torch are installed in that order.  Any other order results
in dependent cuda package clashes.

Resolves: #12

requirements.txt

@@ -1,12 +1,5 @@
 # --- Core training (Microwakeword) ---
 
-# TensorFlow stack (match CUDA/cuDNN in the base image)
-tensorflow==2.17.1
-tensorboard==2.17.1
-tensorboard-data-server==0.7.2
-tensorflow-io-gcs-filesystem==0.37.1
-# tensorflow-estimator not needed (remove)
-
 numpy==1.26.4
 scipy==1.12.0
 librosa==0.10.2.post1
@@ -25,7 +18,6 @@ bitstruct==8.19.0
 
 # --- Piper sample generation ---
 piper-tts>=1.2.0
-onnxruntime-gpu>=1.16.0
 piper-phonemize-cross==1.2.1
 
 # --- Notebook / tooling ---