This change makes errors more informative in case of functions. We now report...

This change makes errors more informative in case of functions. We now report two additional things:
1. The line number/file name of the culprit node inside the function.
2. The function calling stack

For eg, consider the following code:

import tensorflow as tf
tf.enable_eager_execution()

@tf.contrib.eager.defun
def fn3(x):
  return x + 2

@tf.contrib.eager.defun
def fn2(x):
  tf.assert_equal(fn3(x), 3)
  return 2

@tf.contrib.eager.defun
def fn(x):
  return fn2(x)

@tf.contrib.eager.defun
def main(argv):
  x = fn(2)
  print(x)

if __name__ == '__main__':
  tf.app.run()

In this, the error message would contain the following:
InvalidArgumentError: assertion failed: [] [Condition x == y did not hold element-wise:] [x (PartitionedCall:0) = ] [4] [y (assert_equal/y:0) = ] [3]
	 [[node assert_equal/Assert/Assert (defined at filename:line_num) ]]
	 [[StatefulPartitionedCall]]
	 [[StatefulPartitionedCall]] [Op:StatefulPartitionedCall]

Function calling stack:
main -> fn -> fn2

PiperOrigin-RevId: 228593653