[Indic] Recompose Bengali Ya,Nukta

This is a bunch of hacks for now.

Improves Bengali a bit.

src/hb-ot-shape-complex-indic.cc

@@ -407,7 +407,8 @@ hb_ot_shape_normalization_mode_t
 _hb_ot_shape_complex_normalization_preference_indic (void)
 {
   /* We want split matras decomposed by the common shaping logic. */
-  return HB_OT_SHAPE_NORMALIZATION_MODE_DECOMPOSED;
+  /* XXX sort this out after adding per-shaper normalizers. */
+  return HB_OT_SHAPE_NORMALIZATION_MODE_COMPOSED_DIACRITICS;
 }
 
 
@@ -549,7 +550,6 @@ initial_reordering_consonant_syllable (const hb_ot_map_t *map, hb_buffer_t *buff
     if (base < start)
       base = start; /* Just in case... */
 
-
     /* -> If the syllable starts with Ra + Halant (in a script that has Reph)
      *    and has more than one consonant, Ra is excluded from candidates for
      *    base consonants. */

src/hb-unicode.cc

@@ -258,6 +258,14 @@ hb_unicode_compose (hb_unicode_funcs_t *ufuncs,
 		    hb_codepoint_t     *ab)
 {
   *ab = 0;
+  /* XXX, this belongs to indic normalizer. */
+  if ((FLAG (hb_unicode_general_category (ufuncs, a)) &
+       (FLAG (HB_UNICODE_GENERAL_CATEGORY_SPACING_MARK) |
+        FLAG (HB_UNICODE_GENERAL_CATEGORY_ENCLOSING_MARK) |
+        FLAG (HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK))))
+    return false;
+  /* XXX, add composition-exclusion exceptions to Indic shaper. */
+  if (a == 0x09AF && b == 0x09BC) { *ab = 0x09DF; return true; }
   return ufuncs->func.compose (ufuncs, a, b, ab, ufuncs->user_data.compose);
 }