Fixed a bug affecting table mapping in evaluate_result_str()

[mnamysl]

This PR fixes an issue affecting table matching in the evaluate_result_str() function. In particular, the elements of a sequence were removed when iterating over the sequence, resulting in some elements being skipped and matching being incomplete.

To validate that, I ran the evaluation of the ground-truth data against itself, hoping to get the perfect scores. I used the annotations for the trackB2 (modern documents subset) in the ICDAR 2019 cTDaR table competition. I got the following results:

IOU @ 0.6 -
precision: 0.7933181856433561
recall: 0.7933181856433561
f1: 0.7933181856433561
correct: 9427, gt: 11883, res: 11883
IOU @ 0.7 -
precision: 0.7933181856433561
recall: 0.7933181856433561
f1: 0.7933181856433561
correct: 9427, gt: 11883, res: 11883

IOU @ 0.8 -
precision: 0.7933181856433561
recall: 0.7933181856433561
f1: 0.7933181856433561
correct: 9427, gt: 11883, res: 11883
IOU @ 0.9 -
precision: 0.7933181856433561
recall: 0.7933181856433561
f1: 0.7933181856433561
correct: 9427, gt: 11883, res: 11883

After applying the suggested fix, the expected result is returned.

Evaluation of trackB2
IOU @ 0.6 -
precision: 1.0
recall: 1.0
f1: 1.0
correct: 11883, gt: 11883, res: 11883
IOU @ 0.7 -
precision: 1.0
recall: 1.0
f1: 1.0
correct: 11883, gt: 11883, res: 11883

IOU @ 0.8 -
precision: 1.0
recall: 1.0
f1: 1.0
correct: 11883, gt: 11883, res: 11883
IOU @ 0.9 -
precision: 1.0
recall: 1.0
f1: 1.0
correct: 11883, gt: 11883, res: 11883

[alexander-esser]

Used enumerate in the for-loop; in order to know the index within the loop and to remove the erroneous line remaining_tables.remove(rest) # unsafe... should be ok with the break below