Refactor TaylorF2 and IMRPhenomD

[ravioli1369]

This PR add the following:

• Refactor the existing TaylorF2, PhenomD waveforms into torch.nn.Module classes, making use of self.register_buffer to avoid tensor initializations on the CPU.
• The PhenomD has been made into a subclass of TaylorF2 in order to use its functions without too many imports.
• Addition of a constants.py file to keep everything in one place
• Fixing the TaylorF2 phase term used in PhenomD by subtracting off its 3PN term. Refer https://git.ligo.org/lscsoft/lalsuite/-/blob/master/lalsimulation/lib/LALSimIMRPhenomD.c#L397-398.

[ravioli1369]

@deepchatterjeeligo so, apparently it's not as simple as doing a self.qnmdata to access the register buffers, there seems to be a self.buffers and self.named_buffers method to iterate over the buffers. @EthanMarx could you confirm if this is the right way of accessing them? When I had used self.qnmdata_, I got an error saying that parameter is not a part of my class.

[EthanMarx]

@ravioli1369 When you register a buffer it should be accessible as a Class variable. Can you post the error

[deepchatterjeeligo]

I agree with Ethan. E.g.

>>> class Net(torch.nn.Module):
...     def __init__(self):
...         super().__init__()
...         self.register_buffer('mean', torch.as_tensor(1.0))
...     def forward(self, x):
...         return x
... 
>>> net = Net()
>>> net.mean
tensor(1.)

So this should work self.register_buffer('qnm_data_a', QNMData_a)?

[ravioli1369]

@EthanMarx @deepchatterjeeligo, not sure what happened when I tried to access it using self the first time, but now it seems to be working, I've updated the code accordingly.