[MohammadHijjawi97]

EnglishNumberNormalizer.process_words handles point in two cases: if the next word is not a number word, point is emitted as-is; if the next word is a decimal digit (or numeric), a . is appended. When the next word is a number word but not a decimal digit (and, minus, first, dollars, ...) neither case applies and the word is silently dropped:

input	before	after
that was the point and then we left	that was the and then we left	that was the point and then we left
match point minus two	match -2	match point -2
the point first	the 1st	the point 1st
good point dollars	good dollars	good point dollars
two point and	2 and	2 point and

The fix flushes any pending value and emits point, the same way the double/triple branch already handles a non-numeric follower. Decimal cases are unchanged (3 point 14, one point oh five, two point two five dollars still give 3.14, 1.05, $2.25).

Added four asserts to the parametrized test_number_normalizer (they fail without the change); tests/test_normalizer.py, test_tokenizer.py and the non-CUDA test_timing.py tests pass, and black/isort/flake8 at the pre-commit pins are clean.