fix: index_adjustment_after_clean_extra_whitespace maps cleaned index to the wrong original offset

[lntutor]

What's broken

index_adjustment_after_clean_extra_whitespace subtracts the moved distance when it must add it:

def index_adjustment_after_clean_extra_whitespace(index, moved_indices) -> int:
    return int(index - moved_indices[index])

clean_extra_whitespace_with_index_run returns moved_indices where moved_indices[i] is the number of characters removed before cleaned index i (per its docstring), so the original position of a cleaned character is index + moved_indices[index]. Subtracting maps every character following a collapsed whitespace run to the wrong — or a negative — original offset:

text = "ITEM 1.     BUSINESS"
cleaned, moved = clean_extra_whitespace_with_index_run(text)   # 'ITEM 1. BUSINESS'
i = cleaned.index("B")                                          # 8
index_adjustment_after_clean_extra_whitespace(i, moved)        # 4  (a space) -- should be 12

Only + round-trips (text[i + moved[i]] == cleaned[i]) for every kept character, including multi-run cases like "a b c".

Fix

Change the operator to + and add a round-trip regression test. (Distinct from #4166, which optimized the producer clean_extra_whitespace_with_index_run and left this sign bug untouched.)

Tests

Adds test_index_adjustment_maps_cleaned_index_back_to_original. Full test_unstructured/cleaners/test_core.py passes (89).

[cubic-dev-ai]

No issues found across 2 files

<sub>Shadow auto-approve: would auto-approve. Fixes a sign bug in index adjustment mapping; bounded fix with regression test.

Re-trigger cubic</sub>