[S2] normalizer.py no-op replace — smart apostrophes never normalized to ASCII

[erinepshovel-code]

core/parsing/normalizer.py::normalize_text_for_matching contains:

.replace("’", "’")   # U+2019 → U+2019, no-op
.replace("‘", "’")   # U+2018 → U+2019 (curly-open → curly-close, still not ASCII)

ASCII apostrophe (U+0027) is never produced. Bones-map keys for contraction fragments (n't, 'll, 'd, 're, 'm, 've, 's) are written with ASCII '. Input containing smart apostrophes (U+2019 in "don't") survives unchanged, looks up against ASCII-keyed bone entries, and misses — producing incorrect tokenization for any real-world word-processed or copy-pasted text.

Fix:

.replace("’", "'")   # RIGHT SINGLE QUOTATION MARK → ASCII apostrophe
.replace("‘", "'")   # LEFT SINGLE QUOTATION MARK → ASCII apostrophe