Endian fix for Strings in s390x (#4194)

1. The change in StaticString's withUTF8Buffer is to store the UTF-8 code unit properly in the buffer.

2. The change in StringCore's _nthContiguous is to represent the UTF-16 data correctly when it contains the high byte.

We ran validation tests and foundation tests with the change on x86-64 and s390x. No new regression is observed.

Author: Gary Liu

stdlib/public/core/StaticString.swift

@@ -132,7 +132,11 @@ public struct StaticString
       var buffer: UInt64 = 0
       var i = 0
       let sink: (UInt8) -> Void = {
+#if _endian(little)
         buffer = buffer | (UInt64($0) << (UInt64(i) * 8))
+#else
+        buffer = buffer | (UInt64($0) << (UInt64(7-i) * 8))
+#endif
         i += 1
       }
       UTF8.encode(unicodeScalar, into: sink)

stdlib/public/core/StringCore.swift

@@ -299,8 +299,12 @@ public struct _StringCore {
     // Always dereference two bytes, but when elements are 8 bits we
     // multiply the high byte by 0.
     // FIXME(performance): use masking instead of multiplication.
+#if _endian(little)
     return UTF16.CodeUnit(p.pointee)
       + UTF16.CodeUnit((p + 1).pointee) * _highByteMultiplier
+#else
+    return _highByteMultiplier == 0 ? UTF16.CodeUnit(p.pointee) : UTF16.CodeUnit((p + 1).pointee) + UTF16.CodeUnit(p.pointee) * _highByteMultiplier
+#endif
   }
 
   /// Get the Nth UTF-16 Code Unit stored.