Fix heap-buffer-overflow in ConvertingString<> state buffer allocation

On Linux, sizeof(wchar_t) = 4 bytes but sizeof(SQLWCHAR) = 2 bytes
(unixODBC defines SQLWCHAR as unsigned short).  The ConvertingString
constructor used sizeof(wchar_t) to convert a byte-count argument
into the number of narrow characters needed:

    lengthString = length / sizeof(wchar_t);  // = 12/4 = 3 on Linux

For SQLGetDiagRecW the state buffer is declared as State(12, sqlState),
giving lengthString=3 and Alloc() allocating 3+2=5 bytes.  strcpy()
then writes the 5-character SQL state plus its NUL terminator (6 bytes)
into that 5-byte buffer, producing a 1-byte heap-buffer-overflow caught
by AddressSanitizer.

The same latent bug exists in SQLErrorW (same State(12, sqlState) pattern).

Fix: divide by sizeof(SQLWCHAR) instead of sizeof(wchar_t).
sizeof(SQLWCHAR) == 2 on all platforms (Windows: SQLWCHAR=wchar_t=2;
Linux/unixODBC: SQLWCHAR=unsigned short=2), so the formula now yields:

    lengthString = 12 / sizeof(SQLWCHAR) = 6

and Alloc() allocates 6+2=8 bytes, comfortably holding the SQL state.

On Windows sizeof(wchar_t)==sizeof(SQLWCHAR)==2, so this change is
a no-op there.

Found by: AddressSanitizer (introduced in CI via PR #288/#289)
Test: DataTypeTest.SmallintRoundTrip -> ExecIgnoreError -> SQLExecDirect
      -> unixODBC dispatcher -> SQLGetDiagRecW -> sqlGetDiagRec(strcpy)

Author: fdcastel

MainUnicode.cpp

@@ -85,7 +85,7 @@ class ConvertingString
 			if ( length == SQL_NTS )
 				lengthString = 0;
 			else if ( retCountOfBytes )
-				lengthString = length / sizeof(wchar_t);
+				lengthString = length / sizeof(SQLWCHAR);
 			else
 				lengthString = length;
 		}