TST: stats.Normal: bump tolerance on test of logcdf

[mdhaber]

Reference issue

e.g. gh-22274

What does this implement/fix?

There is a test failing in main with the latest release of hypothesis. It looks like it is able to find a little region where quadrature is less accurate than usual:

import numpy as np
from scipy import stats
X = stats.Normal(mu=-0.044622861356271804, sigma=1.234783885368829)
X.logcdf(-0.84)       - X.logcdf(-0.84, method='quadrature')        # np.float64(0.0)
X.logcdf(-0.84693408) - X.logcdf(-0.84693408, method='quadrature')  # np.float64(1.1598066906870486e-07)
X.logcdf(-0.85)       - X.logcdf(-0.85, method='quadrature')        # np.float64(0.0)

So this bumps the tolerance slightly.

[dschmitz89]

Thanks Matt.