CGemv fails with strides0 x

Description

import pytensor
import pytensor.tensor as pt
import numpy as np
from pytensor.compile.mode import get_mode
from pytensor.tensor.blas_c import CGemv

A = pt.tensor("A", shape=(None, None))
x = pt.tensor("x", shape=(None,))

out = CGemv(inplace=False)(
    pt.empty(A.shape[0]),
    pt.constant(np.array(1.0, dtype="float64")),
    A,
    x,
    pt.constant(np.array(0.0, dtype="float64"))
)
fn = pytensor.function([A, x], out)
A_test = np.array(np.ones((5, 3)), dtype="float64")
x_test = np.array(1.0, dtype="float64")
x_test = np.broadcast_to(x_test, (3,))  # x_test.strides == (0,)
fn(A_test, x_test)  # Intel oneMKL ERROR: Parameter 8 was incorrect on entry to DGEMV .

This showed up in a PyMC test now that DimShuffle creates 0-strides inputs

Start at the CGemv entry point in pytensor.tensor.blas_c and reproduce the provided example with the broadcasted x input whose stride is zero. Trace how that input reaches the BLAS call, then verify the example runs without the DGEMV parameter error and add coverage for the zero-stride case.