DynamicLinearRegression fails when specifying `drift_scale_prior`

[connesy]

import tensorflow as tf
from tensorflow_probability.python.sts.internal import util as sts_util
tfd = tfp.distributions

_, observed_stddev, _ = sts_util.empirical_statistics(observed_time_series)
drift_scale_prior = tfd.LogNormal(
    loc=tf.math.log(0.05 * observed_stddev), 
    scale=3.0, 
    name="drift_scale_prior"
)

component = tfp.sts.DynamicLinearRegression(
    design_matrix=design_matrix,
    observed_time_series=observed_time_series, 
    drift_scale_prior=drift_scale_prior,
)

The above rasies UnboundLocalError: local variable 'observed_stddev' referenced before assignment.

This happens when specifying drift_scale_prior for DynamicLinearRegressor, since observed_stddev is only created when drift_scale_prior is None lines 285-290 but is used later on line 305, which is where the above example fails.

tensorflow-probability==0.17.0, tensorflow==2.9.0