[MNT] Test consistency between CPU & GPU version of ROCKET

[baraline]

Describe the issue

After the introduction of the GPU version of ROCKET in #1199, we need new test in order to make sure the CPU and GPU version of rocket are consistent when given the same random seed.

Suggest a potential alternative/fix

The goal of this issue is to implement a test here that will transform a small test dataset using both CPU and GPU versions of ROCKET initialized with the same random seed, and test that the transformation result are similar using the assert_array_almost_equal function from numpy.testing

Additional context

No response

[hadifawaz1999]

given both versions uses numpy to randomly generate the convolution parameters, it should be simple, if it was one numpy one tensorflow then it would have been problematic

we should make sure both versions are using the new api generator in numpy, use the same dataset, and make sure both normalize the input

NOTE: the rocket-gpu only works on equal length series, limitations of accelerating with gpu is batching, and batching only works with equal length series So we should test only with equal length series, univariate and multivariate

[TonyBagnall]

replaced by #1699