Slow extraction after recent PRs

[adbar]

Hi @unsleepy22, I just ran tests on the benchmark with pyinstrument tests/comparison_small.py and there seems to be an issue with your PR which improve the results but have a major cost in terms of timing (say 3-4x as slow overall).

The problem seems to be in determine_returnstring > xmltotxt > process_element. My guess is that a slow test with descendants or ancestors is run recursively although there is no need.

The function is_in_table_cell() is a potential culprit, its use should be limited to cases where it's absolutely necessary and/or there should be a better way to perform the test after all.