Duplicate method definitions in tensor.py

[Terminay]

_graph_nodes(), grad_check(), _eval_forward(), and zero_grad_all() are each defined twice in leanpass/tensor.py (lines 348/466, 365/483, 401/528, 459/577)

The second definitions silently override the first.
The two _eval_forward() implementations are also different, one uses a dict lookup, the other mutates node.data directly.

[github-actions]

The issue highlights that the methods _graph_nodes(), grad_check(), _eval_forward(), and zero_grad_all() are defined twice in leanpass/tensor.py, causing the second definitions to override the first. Notably, the two _eval_forward() implementations differ, with one using a dictionary lookup and the other directly mutating node.data. This redundancy could lead to unexpected behavior or maintenance challenges.