Fix #418: add tests for padded_to_ragged2d

[mhucka]

This adds some simple tests for an untested function intfq_utility_ops_test.py.

[gemini-code-assist]

Code Review

This pull request updates tfq_utility_ops_test.py by correcting a typo, adding more test cases to test_padded_to_ragged, and implementing a new test for padded_to_ragged2d. Feedback indicates that the logic in test_padded_to_ragged2d contains an error in tensor slicing that only works for square matrices and suggests that the test implementation is too similar to the actual code, potentially missing bugs.

[gemini-code-assist]

The logic used to calculate the expected value in test_padded_to_ragged2d is flawed and only works for square matrices.

1. Incorrect Slicing: tensor_arr[:, 0] selects the first row of each matrix in the batch, resulting in a mask of shape [batch, columns]. When used in tf.ragged.boolean_mask(tensor_arr, col_mask), it attempts to mask the second dimension (rows) of tensor_arr. This will cause a shape mismatch error if the number of rows does not equal the number of columns. To correctly mask rows based on their first element, you should use tensor_arr[:, :, 0].
2. Tautological Test: The test currently replicates the implementation of the op it is testing. This means that if the op contains a logic error (which it does, as it uses the same incorrect slicing), the test will still pass, failing to catch the bug.

Note that the implementation of padded_to_ragged2d in tfq_utility_ops.py likely requires a similar fix.

Suggested change

	col_mask = tf.abs(tensor_arr[:, 0]) < 1.1
	masked = tf.ragged.boolean_mask(tensor_arr, col_mask)
	mask = tf.abs(masked) < 1.1
	expected = tf.ragged.boolean_mask(masked, mask)
	row_mask = tf.abs(tensor_arr[:, :, 0]) < 1.1
	masked_rows = tf.ragged.boolean_mask(tensor_arr, row_mask)
	element_mask = tf.abs(masked_rows) < 1.1
	expected = tf.ragged.boolean_mask(masked_rows, element_mask)