build profiling image in CI, add arm64 support to AWS notebooks

[jameslamb]

More miscellaneous refactoring.

• omits the windows/ directory from LightGBM (no longer necessary to build the package, as of [python-package] remove MSVS solution files from sdist microsoft/LightGBM#6698)
• adds profiling image to CI
• explicitly sets CPU architecture in dask_cloudprovider.FargateCluster example, to fix this:
• mounts in LightGBM/ as read-write at build time

Notes for Reviewers

Why set CPU architecture?

This repo only builds single-architecture container images.
The notebooks are intended to be run locally, from the same machine where you built all the images.

dask_cloudprovider.FargateCluster defaults to x86_64.
When I built the images on my M2 Mac (which is arm64) then tried to run the demo-aws notebook as-is, I saw this error on ECS:

exec /usr/local/bin/dask-scheduler: exec format error

This fixes that by detecting CPU architecture on the machine where the notebook is running, and passing that through. Thanks to @dmitry-livchak for adding that support in dask/dask-cloudprovider#425 😁

Why make LightGBM/ read-write at build time?

Without it, the build fails like this:

cp: cannot create directory './lightgbm-python': Read-only file system

Because build-python.sh from LightGBM creates a temporary directory and moves files into it (code link).