[lagru]

This bundles a few things that I should have probably committed separately while iterating on them:

1. Closes Include shape of NumPy array #76. Shape information in doctypes is now translated into annotations. This even goes for shapes with optional dimensions – which use expansion like in the plugin example to create a union of arrays – one for each supported dimensionality. Shapes with "..." are treated as unbounded dimensionality and translated to tuple[Any, ...] like NumPy v2.5 suggests. Empty shapes are supported too.

2. Dtype unions are now also expanded into a union of arrays each with one of the dtypes. This is because I'm not sure if np.dtype actually supports unions. I can't find examples of that in NumPy's repo. For now, expansion should be on the safe side.

3. Closes ArrayLike limitations #125. Shape and dtype are now stripped from ArrayLikes which don't support them as their are a union that can't be subscribed. This means, ArrayLikes are a really generic broad type.

4. Cleans up some issues with the grammar: the literal and subscriptions rules have some necessary overlap. The subscription rule should now win unless the expression contains an actual literal type.

Also some refactoring, and especially new stuff like expand_expression, Expr.{find_all|try_find_one|find_one} enable all this.

Release note

For maintainers and optionally contributors, please refer to changelist's README on how to document this PR for the release notes.

Experimental: Include shape information when translating doctypes
to annotations with `numpy.typing.NDArray`. This includes support
for shapes with optional dimensions which use expansion like in
the plugin example to create a union of arrays – one for each
supported rank. Shapes with "..." are treated as an unspecified
number of dimensions and are to `tuple[Any, ...]`. Empty shapes
are supported too.
{label="enhancement"}

Expand arrays whose dtype is a union of dtypes into a union of
arrays, one for each given dtype. This might be necessary if
`numpy.typing.dtype` does not support unions as arguments.
{label="fix"}

Strip shape and dtype from *ArrayLikes* when converting them
to an annotation since `numpy.typing.ArrayLike` doesn't support
adding that information.
{label="fix"}