[USE] Remove non-canonical decompositions

We have had added this in Indic shaper to assist shaping these scripts.
In Universal Shaping Engine however, it is up to font designer to
decompose them.  Hence moving them from Indic shaper to USE was
wrong.

Fixup for f6ba63b2

Part of fixing https://github.com/behdad/harfbuzz/issues/387

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

@@ -572,28 +572,6 @@ decompose_use (const hb_ot_shape_normalize_context_t *c,
      */
     case 0x1112Eu : *a = 0x11127u; *b= 0x11131u; return true;
     case 0x1112Fu : *a = 0x11127u; *b= 0x11132u; return true;
-
-    /*
-     * Decompose split matras that don't have Unicode decompositions.
-     */
-
-    /* Limbu */
-    case 0x1925u  : *a = 0x1920u; *b= 0x1923u; return true;
-    case 0x1926u  : *a = 0x1920u; *b= 0x1924u; return true;
-
-    /* Balinese */
-    case 0x1B3Cu  : *a = 0x1B42u; *b= 0x1B3Cu; return true;
-
-#if 0
-    /* Lepcha */
-    case 0x1C29u  : *a = no decomp, -> LEFT; return true;
-
-    /* Javanese */
-    case 0xA9C0u  : *a = no decomp, -> RIGHT; return true;
-
-    /* Sharada */
-    case 0x111BFu  : *a = no decomp, -> ABOVE; return true;
-#endif
   }
 
   return (bool) c->unicode->decompose (ab, a, b);