Allow specifying amount of boundary condition test points

[g-w1]

Something similar can be done in data/pde_operator.py and I may do this in a future pull request.

Should I add num_test_boundary to the examples? I'm not sure if this is a problem in the real-world. It seems to me that doing this would prevent (or at least detect) overfitting to the boundary conditions selected.

I also fixed a trivial TODO in pde.py (rename train_x_all to train_x_pde). Let me know if you want me to split this into a seperate PR.

[lululxvi]

I also fixed a trivial TODO in pde.py (rename train_x_all to train_x_pde). Let me know if you want me to split this into a seperate PR.

Yes, too many changes in this PR. Please split.

[g-w1]

Yes, too many changes in this PR. Please split.

I have split it into #1113 .

[lululxvi]

There might be some classes that also use num_test. You need double check.

[lululxvi]

Another main issue is that in losses, we use self.num_bcs for both training and testing. If you generate test boundary points separately, then self.num_bcs doesn't work for test points.

[g-w1]

I will modify ide.py and fpde.py to add num_test_domain and num_test_boundary and will create self.num_bcs_train and self.num_bcs_test. Thank you for the feedback.