.take after .shuffle doesn't work as expected

Performing .take(1).repeat() after .shuffle(N) returns N different elements.

Reproduction instructions

dataset = tf.data.Dataset.from_tensor_slices([1, 2, 3, 4])
dataset = dataset.shuffle(4)
dataset = dataset.take(1)
dataset = dataset.repeat()
dataset = dataset.take(10)
for x in dataset:
  print(x)

This will print 10 different, random integers in the range 1 - 4.

Expected behavior
I would have expected this to print only a single unique integer. It appears the early .shuffle interferes with the later .take, which is very counterintuitive.

See this colab:

https://colab.research.google.com/drive/1UYu3TyzMd2nPs19YyNrViFQZxfSyw8WR?usp=sharing

Start by running the Python reproduction with tf.data.Dataset, shuffle(4), take(1), repeat(), and take(10), then compare the output with the linked Colab. Trace how these chained operations interact and confirm that the completed behavior yields one repeated value rather than ten different elements.