Fix ProgressBoard AttributeError and unconditional label assignment in torch/tensorflow/mxnet backends

[Chessing234]

Bug

In ProgressBoard.draw() across the PyTorch, TensorFlow, and MXNet backends, the label-setting code is:

if not self.xlabel: self.xlabel = self.x
axes.set_xlabel(self.xlabel)
axes.set_ylabel(self.ylabel)

Two problems:

1. self.x is not an attribute of ProgressBoard — it is never stored on the instance. When xlabel is None (the default) or an empty string, this line raises AttributeError: 'ProgressBoard' object has no attribute 'x'.

2. The truthiness check if not self.xlabel treats an explicitly-supplied empty string "" as equivalent to None, unintentionally overwriting it.

Root cause

The fallback self.xlabel = self.x is a dead code remnant; x is a local parameter of draw() not a stored attribute. When xlabel is None the intent is simply to omit the axis label, not to substitute the numeric x-coordinate.

The same bug was fixed in d2l/jax.py by PR #2704/#2705 but those PRs only touched the JAX backend.

Fix

Guard both set_xlabel and set_ylabel with explicit is not None checks (matching the JAX fix) in all three remaining backends: torch.py, tensorflow.py, and mxnet.py.