Fix the length returned by UTF-32 conversion for non-BMP input

Don't optimize the required length as this is a tiny gain resulting in big
problems with the strings containing surrogates for which the actual result is
shorter than the length returned, resulting in extra NUL bytes at the end of
the converted buffer.

This is similar to 3410aa372f (see #16298) but
for UTF-32 and not UTF-16.

Closes #17070.

tests/strings/unicode.cpp

@@ -403,6 +403,19 @@ void UnicodeTestCase::ConversionUTF16()
     wxMBConvUTF16BE().cMB2WC("\xd8\x03\xdc\x01\0" /* OLD TURKIC LETTER YENISEI A */, wxNO_LEN, &len);
     CPPUNIT_ASSERT_EQUAL( 1, len );
 #endif // UTF-32 internal representation
+
+#if SIZEOF_WCHAR_T == 2
+    // Verify that the length of UTF-32 string is correct even when converting
+    // to it from a longer UTF-16 string with surrogates.
+
+    // Construct CAT FACE U+1F431 without using \U which is not supported by
+    // ancient compilers and without using \u with surrogates which is
+    // (correctly) flagged as an error by the newer ones.
+    wchar_t ws[2];
+    ws[0] = 0xd83d;
+    ws[1] = 0xdc31;
+    CPPUNIT_ASSERT_EQUAL( 4, wxMBConvUTF32BE().FromWChar(NULL, 0, ws, 2) );
+#endif // UTF-16 internal representation
 }
 
 void UnicodeTestCase::ConversionUTF32()