simd: fix MSVC ARM64 build with SJPEG_NEON16_OP/8_OP

[skal65535]

• MSVC aliases all NEONxx_t types to one __n128/__n64 union
• => duplicate operator overloads, C2535
• skip extra conversions for MSVC (not clang-cl, distinct types there)

[jzern]

Let me verify this before submitting.

[jzern]

Closer, now the same thing for Store16_NEON:

src\simd.h(192,1): error C2084: function 'void Store16_NEON(uint8x16_t,void *)' already has a body

[jzern]

Closer, now the same thing for Store16_NEON:

src\simd.h(192,1): error C2084: function 'void Store16_NEON(uint8x16_t,void *)' already has a body

Updating that to match passes. Note I also have the following addition to enable NEON. Noticed the failure when I was getting ready to push the patch.

diff --git a/src/simd.h b/src/simd.h
index 2636e41..bfb892d 100644
--- a/src/simd.h
+++ b/src/simd.h
@@ -34,7 +34,9 @@
 #define SJPEG_USE_AVX2
 #endif

-#if !defined(SJPEG_USE_NEON) && (defined(__ARM_NEON__) || defined(__aarch64__))
+#if !defined(SJPEG_USE_NEON) &&                                          \
+    (defined(__ARM_NEON__) || defined(__aarch64__) || defined(_M_ARM) || \
+     defined(_M_ARM64) || defined(_M_ARM64EC))
 #define SJPEG_USE_NEON
 #endif

[jzern]

clang-cl builds with the above changes.

[skal65535]

LGTM