fix: Apply scale parameter within exp in MultiScaleKernel

[Markus28]

It appears that there is a typo in the implementation of MultiScaleKernel. Namely, @pjhartout and I would expect that the scale parameter self.sigma should be applied before exponentiation in these two lines:

pytorch-topological/torch_topological/nn/multi_scale_kernel.py

Line 134 in aba1474

M = torch.exp(-nom) / (8 * self.sigma)

pytorch-topological/torch_topological/nn/multi_scale_kernel.py

Line 135 in aba1474

M -= torch.exp(-denom) / (8 * self.sigma)

For reference, consider equation 10 in https://arxiv.org/pdf/1412.6821.
In the current implementation, self.sigma only acts as a multiplicative factor on the kernel and thereby does not meaningfully change its behavior.

This PR fixes the two lines above. The tests are passing locally for me.

[Pseudomanifold]

Looks very good to me, thanks for your contribution!