glm: drop register storage class from type_half.inl

toFloat16 declares its sign/exponent/significand locals without a
storage class specifier, so the header parses under C++17. Clang
rejects register outright at that standard, which puts the macOS jobs
on the same footing as the GCC targets.

glm/core/type_half.inl

@@ -135,9 +135,9 @@ namespace detail
 		// of float and half (127 versus 15).
 		//
 
-		register int s =  (i >> 16) & 0x00008000;
-		register int e = ((i >> 23) & 0x000000ff) - (127 - 15);
-		register int m =   i        & 0x007fffff;
+		int s =  (i >> 16) & 0x00008000;
+		int e = ((i >> 23) & 0x000000ff) - (127 - 15);
+		int m =   i        & 0x007fffff;
 
 		//
 		// Now reassemble s, e and m into a half: