FilteredBackProjection crashes on any input

Hi, thanks for maintaining PyTomography - I've been learning XCT reconstruction algorithms and reading the source has been really helpful.

I tried running FilteredBackProjection on a small SPECT example and it always crashes. Looking at algorithms/fbp.py, three things stand out:

1. __init__ takes projections but never stores it (self.proj), while __call__ reads self.proj -> AttributeError on the first use.

2. The FFT calls use axis=-2, but torch.fft.fft / ifft take dim (in older torch versions axis is not accepted at all, and the project requires torch>=1.10.2).

3. len(self.system_matrix.proj_meta.shape[0]) calls len() on an int (shape[0] is the number of projections) -> TypeError. Looks like it should just be torch.pi / self.system_matrix.proj_meta.shape[0].

I confirmed the first crash locally (AttributeError on self.proj). It's possible I'm missing something, but this class seems unusable on any input right now. Let me know if you'd like me to open a PR for the fixes.

Start in algorithms/fbp.py and inspect FilteredBackProjection.init and call. Reproduce the small SPECT example, then verify the three reported failures involving self.proj, FFT dimensions, and projection-count handling are resolved. Done means FilteredBackProjection can be called on the example without these AttributeError or TypeError crashes.