Remove unnecessary double reinterpret cast in blend call

As pointed out by @zkingston, sign_mask_sin is already float32x4_t
and blend() expects float32x4_t, so no conversion is needed.

The blend function internally handles the conversion to uint32x4_t
for vbslq_f32, making the double cast redundant.

Author: saahu27

src/impl/vamp/vector/neon.hh

@@ -602,7 +602,7 @@ namespace vamp
 
             auto poly_mask_f = IntVector::template as<VectorT>(poly_mask);
             auto ys = blend(y2, y1, poly_mask_f);
-            return blend(ys, neg(ys), vreinterpretq_f32_u32(vreinterpretq_u32_f32(sign_mask_sin)));
+            return blend(ys, neg(ys), sign_mask_sin);
         }
 
         // NOTE: Dummy parameter because otherwise we get constexpr errors with set1_ps...