The operations peformed by the stubless code should roughly match what 
operations MIDL outputs in code, and it doesn't do this.

We're never going to implement the undocumented ones and we're moving 
more and more towards the architecture of the more recent versions of 
native rpcrt4.

Spotted by Ken Thomases.

Convert ex->cb into a character count before assigning it to nCount.

Remove some dubious minus ones that would result in different character 
counts for the Unicode and ANSI paths.

riched20: The wParam parameter to WM_GETTEXT contains the number of characters, not the number of bytes.

Fix up some places in the WM_GETTEXT handler where it was assumed that 
it was a byte count.

Based on a patch by Kris Moore.

Spotted by Christoph von Wittich <Christoph@ApiViewer.de>.