fix: preserve paragraph breaks as sentence boundaries in normalizer

[Christian-Sidak]

Summary

• The text normalizer was replacing all newlines (including \n\n paragraph breaks) with spaces, causing headings and body text to be merged into one continuous phrase with no prosodic pause.
• Double-newline paragraph breaks are now converted to a period before the newline-to-space pass, giving the TTS engine a sentence boundary to pause on.
• If the character preceding the paragraph break is already sentence-ending punctuation (., !, ?), no extra period is inserted, avoiding double punctuation.
• Single newlines continue to become spaces as before.

Failure case from the issue:

Input:  "Steven Erikson\n\nFive riders drew rein in the pass."
Before: "Steven EriksonFive riders drew rein in the pass."  (merged, no pause)
After:  "Steven Erikson. Five riders drew rein in the pass."  (natural boundary)

Partial resolution of #519

Test plan

• test_paragraph_breaks_preserved_as_sentence_boundary: exact failure case from the issue report
• test_paragraph_break_no_double_punctuation: paragraph already ends with ., !, or ?
• test_single_newline_becomes_space: soft line breaks still collapse to a space
• test_multiple_blank_lines_treated_as_paragraph_break: three or more consecutive newlines produce exactly one period
• All existing normalizer tests pass without modification

[remsky]

Glad to merge in for the stated case fix, though I think to fully close #519 , the chunker needs to split at blank lines too for normalize: false to respect the paragraph breaks. The UAX change on PR #520 changed the flow a bit.

refactor/normalization-classes is a more general fix in the same line if you spot anything