Fix NoneType crash when model.eval() does not return self

[Pchambet]

Problem

Passing a model whose .eval() does not return self causes an immediate crash in BaseCAM.__init__:

  File "pytorch_grad_cam/base_cam.py", line 28, in __init__
    self.device = next(self.model.parameters()).device
                       ^^^^^^^^^^^^^^^^^^^^^
AttributeError: 'NoneType' object has no attribute 'parameters'

Root cause

self.model = model.eval()  # None if eval() doesn't return self

Standard nn.Module.eval() returns self, but some third-party wrappers (quantization toolkits, custom tracing wrappers) override .eval() without propagating the return value — making model.eval() return None.

Fix

Separate assignment from the .eval() call:

self.model = model
self.model.eval()

This guarantees self.model always holds the model reference, regardless of what .eval() returns.

Precedent in this codebase

guided_backprop.py already uses the safe pattern:

# guided_backprop.py, line 49
self.model.eval()

This PR aligns base_cam.py with that existing convention.

Fixes #569