Fix major CI bugs

[danielholanda]

Description

This PR solves 1 major and 1 minor CI bug:

• [BUG 1] TKML was testing the pypi version rather than the current version (Major)
• [BUG 2] TKML timeout fails on main (minor)

Bug 1

TKML was testing the pypi version rather than the current version. This was effectively causing all of the plugin and turnkey tests to be false positives.

This was caused by hardcoding the supported version of TKML into the plugins setup.py AND installing plugins after installing tkml itself (overwriting this).

This issue has been solved by carefuly changing the installation order and being more flexible on the versions of tkml that the plugins accept.

Bug 2

Pytorch export significantly improved recently, causing our timeout test to behave as shown below:

Info: Running turnkey on ~\turnkeyml\src\turnkeyml\common\generated\test_corpus\extras\timeout.py

Building "timeout"
    + Discovering PyTorch models   
    x Exporting PyTorch to ONNX   
      Optimizing ONNX file   

timeout.py:
        model (executed 1x - 0.16s)
                Location:       C:\\Users\\danie\\turnkeyml\\src\\turnkeyml\\common\\generated\\test_corpus\\extras\\timeout.py, line 22
                Parameters:     500,001,000 (1.86 GB)
                Input Shape:    'x': (500000,)
                Build dir:      C:\Users\danie\turnkeyml\src\turnkeyml\common\generated\cli_cache_dir\timeout
                Status:         Error: The serialized model is larger than the 2GiB limit imposed by the protobuf library. Therefore the output file must be a file path, so that the ONNX external data can be written to the same directory. Please specify the output file name.
(...)
    _C._jit_pass_onnx_graph_shape_type_inference(
RuntimeError: The serialized model is larger than the 2GiB limit imposed by the protobuf library. Therefore the output file must be a file path, so that the ONNX external data can be written to the same directory. Please specify the output file name.

Instead of relying on an extremely long export, we simply make the model take long to be discovered by adding a busy wait using sleep. As a result, we can reliably test our timeout feature.

[ramkrishna2910]

Looks good!