feat: rule-based sentence segmentation — fix abbreviation splitting

Replace unicode_segmentation::unicode_sentences() which splits on
abbreviation periods (Mr., p.m., H.G., etc.), inflating 1-2 word
sentence bin to 22% and deflating mean to 6.8 words.

New forward-scanning splitter with:
- 3-tier abbreviation classification (strong/internal/weak)
- Mid-dotted-sequence detection (e.g., p.m., H.G., B.B.C.)
- Decimal number handling (42.50)
- Ellipsis handling (split only before uppercase)
- Dialog tag detection ("Go!" he said — no split at !)
- 20 unit tests covering all edge cases

Fingerprint after fix: mean 6.8→12.6, 1-2 bin 21.9%→9.8%,
long sentences (18+ words) 0%→22.3%, p95 14→35.

Author: longevityboris

src/stylometry/features/lengths.rs

@@ -39,8 +39,8 @@ pub const HISTOGRAM_BIN_CENTERS: &[f64] =
 pub fn sentence_length_histogram(text: &str) -> Vec<f64> {
     use unicode_segmentation::UnicodeSegmentation;
 
-    let lengths: Vec<u32> = text
-        .unicode_sentences()
+    let lengths: Vec<u32> = super::sentences::split_sentences(text)
+        .iter()
         .map(|s| s.unicode_words().count() as u32)
         .filter(|&l| l > 0)
         .collect();
@@ -126,8 +126,8 @@ pub fn word_lengths(text: &str) -> LengthStats {
 }
 
 pub fn sentence_lengths(text: &str) -> LengthStats {
-    let lengths: Vec<f64> = text
-        .unicode_sentences()
+    let lengths: Vec<f64> = super::sentences::split_sentences(text)
+        .iter()
         .map(|s| s.unicode_words().count() as f64)
         .filter(|&l| l > 0.0)
         .collect();

src/stylometry/features/mod.rs

@@ -13,4 +13,5 @@ pub mod ngrams;
 pub mod punctuation;
 pub mod readability;
 pub mod richness;
+pub mod sentences;
 pub mod vocabulary;

src/stylometry/features/readability.rs

@@ -22,7 +22,7 @@ pub struct ReadabilityStats {
 impl ReadabilityStats {
     pub fn compute(text: &str) -> Self {
         let words: Vec<&str> = text.unicode_words().collect();
-        let sentences = text.unicode_sentences().count() as f64;
+        let sentences = super::sentences::split_sentences(text).len() as f64;
         let word_count = words.len() as f64;
 
         if word_count == 0.0 || sentences == 0.0 {