TransformedVariable does not follow array convention for equality (==) comparison

A tfp.util.TransformedVariable generally behaves like any tensor-like object. For example, a less-than-equal comparison works element-wise, as expected:

import tensorflow as tf
import tensorflow_probability as tfp
var = tf.Variable(1.0)
assert var <= 1.0 and var >= 1.0 and var == 1.0  # passes
t_var = tfp.util.TransformedVariable(1.0, tfp.bijectors.Identity())
assert t_var <= 1.0 and t_var >= 1.0  # passes

However, equality comparison does not behave likewise, and the following fails:

assert t_var == 1.0  # FAILS

This is completely unexpected behaviour which can lead to very subtle downstream bugs.

The reason for this behaviour seems to be due to explicitly removing __eq__ and __ne__ in https://github.com/tensorflow/probability/blob/a91efeb17a6f6fdae97d3465878f5fc7fba12d30/tensorflow_probability/python/util/deferred_tensor.py#L86

What was the motivation for excluding equality comparisons? Would there be any objections to a PR that removes that line, thereby allowing tfp TransformedVariables to behave more expectedly like tf Variables?

Open tensorflow_probability/python/util/deferred_tensor.py at line 86 and inspect why eq and ne are removed. Run the Python reproducer from the issue and compare TransformedVariable equality with TensorFlow Variable behavior. Done means equality comparisons behave consistently without breaking the existing tensor-like comparisons.