Optional tensorflow dependency

• I've noticed that TensorFlow is not used neither in tests nor in notebooks explicitly, made it an optional dependency. I think the only place it appears is wake steering notebook which installed surrogates interface, which will bring in TensorFlow anyhow. There is also something in loads.py but if required then user can install this manually by specifying pip install topfarm[tensorflow].
• Updated docs to reflect the change.
• Removed docker/ directory as it's not used any more.
• Couple of trivial changes to CI config, left over misconfiguration from recent changes...
• Also, Windows runner was failing on checkout of repository. This is something to do with permission and file locks from previous runs. I simply added, some arguments to git clean and save .pixi directory from run to run (this will also speed up environment recreation) but manual topfarm install is needed then to update the version file.
• Disabled the test_maxiter_CostModelComponent test as it's causing problems in CI; that counter depends on a time spent executing evaluation and is very unpredictable to test for... We could think of implementing a deterministic counter if it's of interest.
• Added a test for __version__ availability and format.

@mikf if the tests pass, and you agree with the change, would be nice to merge this : ) Should be ready right away...