Avoid unaligned load in av1_filter_intra_predictor_neon

Use explicit unaligned byte loads instead of vld1_lane_u32 to avoid
undefined behavior on misaligned above pointer addresses.

Change-Id: I7f717b3012efeb3101899055db54e1e1dd2a557a

av1/common/arm/reconintra_neon.c

diff --git a/av1/common/arm/reconintra_neon.c b/av1/common/arm/reconintra_neon.c
index 95503aa..9ac013f 100644
--- a/av1/common/arm/reconintra_neon.c
+++ b/av1/common/arm/reconintra_neon.c
@@ -115,7 +115,7 @@
 
     int c = 0;
     do {
-      uint8x8_t s1234 = load_u8_4x1(above + c);
+      uint8x8_t s1234 = load_unaligned_u8_4x1(above + c);
       uint8x8_t s1 = vdup_lane_u8(s1234, 0);
       uint8x8_t s2 = vdup_lane_u8(s1234, 1);
       uint8x8_t s3 = vdup_lane_u8(s1234, 2);