pt.tensor("x") without shape= gives misleading error — default implies optional when it isn't

The problem

A user who calls `pt.tensor("x")` without a `shape` argument gets:

TypeError: 'NoneType' object is not iterable

This is opaque. The user hasn't passed `None`, they've omitted an argument that appears optional. The function signature says `shape: ... | None = None`, which tells them "this is optional." It isn't — `shape` is effectively required, because `TensorType` has no way to infer dimensionality without it.

The same applies directly:

>>> TensorType("float64")
TypeError: 'NoneType' object is not iterable

Where

`TensorType.init` iterates over `shape` unconditionally:

`type.py:121-123`

if isinstance(shape, int):
    shape = (shape,)
self.shape = tuple(parse_bcast_and_shape(s) for s in shape)   # shape=None → crash

There's a guard for `int`, but no guard for `None`. The default flows straight into the loop.

Contrast

`pt.xtensor("x", dims=("a",))` works without `shape` because `XTensorType` knows `ndim` from `dims` and fills a default. `TensorType` has no equivalent — which is fine — but the error should reflect that.

What's confusing

1. The signature is misleading: `shape: ... | None = None` declares `None` as the default and a valid type. It's neither.
2. The error says nothing useful: "'NoneType' object is not iterable" tells users nothing about what they did wrong or how to fix it.
3. No pointer to helpers: the error doesn't mention `shape=` or the shorthand constructors.

Proposed fix

Remove the default from `shape` in both `tensor()` and `TensorType.init`, and drop `| None` from the type hints. Python then gives its native message:

# tensor()
def tensor(name=None, *, dtype=None, shape, **kwargs):
    ...

# TensorType.__init__
def __init__(self, dtype, shape, name=None, broadcastable=None):
    ...

Result:

>>> pt.tensor("x")
TypeError: tensor() missing 1 required keyword-only argument: 'shape'

>>> TensorType("float64")
TypeError: TensorType.__init__() missing 1 required positional argument: 'shape'

Zero new code, zero breakage (every call site in the repo already passes `shape=`), and the deprecated `broadcastable` pathway still works since `broadcastable` has its own default and gets assigned to `shape` before use.

Start with TensorType.init at pytensor/tensor/type.py:121-123, then locate the tensor() entry point and inspect how both signatures handle shape. Verify the existing call sites described in the issue and confirm that omitted shape produces Python's missing-argument errors while the deprecated broadcastable pathway remains supported.