Move immintrin/arm_neon includes to where they are used.

src/opts/SkNx_neon.h

 /*
  * Copyright 2015 Google Inc.
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #ifndef SkNx_neon_DEFINED
 #define SkNx_neon_DEFINED
 
+#include <arm_neon.h>
+
 #define SKNX_IS_FAST
 
 // ARMv8 has vrndmq_f32 to floor 4 floats.  Here we emulate it:
 //   - roundtrip through integers via truncation
 //   - subtract 1 if that's too big (possible for negative values).
 // This restricts the domain of our inputs to a maximum somehwere around 2^31.  Seems plenty big.
 static inline float32x4_t armv7_vrndmq_f32(float32x4_t v) {
     auto roundtrip = vcvtq_f32_s32(vcvtq_s32_f32(v));
     auto too_big = vcgtq_f32(roundtrip, v);
     return vsubq_f32(roundtrip, (float32x4_t)vandq_u32(too_big, (uint32x4_t)vdupq_n_f32(1)));
@@ skipping 419 matching lines @@
 
 template<> inline Sk4h SkNx_cast<uint16_t, uint8_t>(const Sk4b& src) {
     return vget_low_u16(vmovl_u8(src.fVec));
 }
 
 template<> inline Sk4b SkNx_cast<uint8_t, uint16_t>(const Sk4h& src) {
     return vmovn_u16(vcombine_u16(src.fVec, src.fVec));
 }
 
 #endif//SkNx_neon_DEFINED