optimize neon loadu_128/storeu_128

vld1q_u8 and vst1q_u8 has no alignment requirements.

This improves performance on Oracle Cloud's VM.Standard.A1.Flex by 1.15% on a 16*1024 input,
 from 13920 nanoseconds down to 13800 nanoseconds (approx)

ref BLAKE3-team/BLAKE3#384

ext/hash/blake3/patches.diff

@@ -54,3 +54,27 @@ index af6c3dadc7..af3bf17bbe 100644
 
  void blake3_compress_in_place(uint32_t cv[8],
                                const uint8_t block[BLAKE3_BLOCK_LEN],
+diff --git a/ext/hash/blake3/upstream_blake3/c/blake3_neon.c b/ext/hash/blake3/upstream_blake3/c/blake3_neon.c
+index 8a818fc78f..c4b4548edf 100644
+--- a/ext/hash/blake3/upstream_blake3/c/blake3_neon.c
++++ b/ext/hash/blake3/upstream_blake3/c/blake3_neon.c
+@@ -9,15 +9,13 @@
+ #endif
+
+ INLINE uint32x4_t loadu_128(const uint8_t src[16]) {
+-  // vld1q_u32 has alignment requirements. Don't use it.
+-  uint32x4_t x;
+-  memcpy(&x, src, 16);
+-  return x;
++    // https://github.com/BLAKE3-team/BLAKE3/pull/384
++    return vreinterpretq_u32_u8(vld1q_u8(src));
+ }
+
+ INLINE void storeu_128(uint32x4_t src, uint8_t dest[16]) {
+-  // vst1q_u32 has alignment requirements. Don't use it.
+-  memcpy(dest, &src, 16);
++    // https://github.com/BLAKE3-team/BLAKE3/pull/384
++    vst1q_u8(dest, vreinterpretq_u8_u32(src));
+ }
+
+ INLINE uint32x4_t add_128(uint32x4_t a, uint32x4_t b) {

ext/hash/blake3/upstream_blake3/c/blake3_neon.c

@@ -9,15 +9,13 @@
 #endif
 
 INLINE uint32x4_t loadu_128(const uint8_t src[16]) {
-  // vld1q_u32 has alignment requirements. Don't use it.
-  uint32x4_t x;
-  memcpy(&x, src, 16);
-  return x;
+    // https://github.com/BLAKE3-team/BLAKE3/pull/384
+    return vreinterpretq_u32_u8(vld1q_u8(src));
 }
 
 INLINE void storeu_128(uint32x4_t src, uint8_t dest[16]) {
-  // vst1q_u32 has alignment requirements. Don't use it.
-  memcpy(dest, &src, 16);
+    // https://github.com/BLAKE3-team/BLAKE3/pull/384
+    vst1q_u8(dest, vreinterpretq_u8_u32(src));
 }
 
 INLINE uint32x4_t add_128(uint32x4_t a, uint32x4_t b) {