support string dims in `as_xtensor`

[williambdean]

Using pymc.dims, using as_tensor is suggested with pymc.dims. However, dims work different with as_xtensor than pmd.<DistributionName>.

import pymc as pm
import pymc.dims as pmd
from pytensor.xtensor import as_xtensor

observed = [5, 7]

coords = {"repo": ["pymc", "pymc-marketing"]}
with pm.Model(coords=coords) as model:
    lam = pmd.HalfNormal("lam", dims="repo")
    pmd.Exponential(
        "obs",
        lam=lam,
        observed=observed,
        dims="repo",
    )

ValueError: Variable [5, 7] must have dims associated with it.
To avoid subtle bugs, PyMC does not make any assumptions about the dims of parameters.
Use `as_xtensor` with the `dims` keyword argument to specify the dims explicitly.

However, a likely use of as_xtensor doesn't act the way pymc supports dims

observed = as_xtensor([5, 7], dims="repo")

coords = {"repo": ["pymc", "pymc-marketing"]}
with pm.Model(coords=coords) as model:
    lam = pmd.HalfNormal("lam", dims="repo")
    pmd.Exponential(
        "obs",
        lam=lam,
        observed=observed,
        dims="repo",
    )

ValueError: Shape (2,) must have the same length as dims ('r', 'e', 'p', 'o')

[ricardoV94]

We may also want to check if functions in xtensor systematically consider a single string as (string,) when multiple are allowed. Even if we deviate from xarray, sounds like the best UX