When handling an exception, NtContinue can be called from within the signal
handler, in which case the raise(SIGUSR2) call ends up getting eaten and
integer register context never ends up getting restored. Switch to the
method used on X86 to avoid these issues.

This currently works only because refcount field is at the same offset between structures.

The libraries used are now staticly linked from the libs directory.

Dumping variable address by itself is barely useful.
Signed-off-by: David Kahurani <k.kahurani@gmail.com>

Without this, wineboot would bail out on ARM64 in the case where x86_64 WOW is
not available but i386 is. Leaving an incomplete prefix without any clear error.

This partially reverts c5daf6a3 (riched20: Call ME_UpdateRepaint
instead of editor_ensure_visible in set_selection.)

Fixes: c5daf6a3

Today, CreateTextServices() returns an Rich Edit object without row
start and end marks, which are expected to exist by many Rich Edit
operations as well as EM_* message handlers.

This leads to a crash when certain messages (e.g., EM_SCROLLCARET) are
sent to the Rich Edit object via ITextServices::TxSendMessage(), unless
ME_WrapMarkedParagraphs() has been called beforehand.

Fix this by calling wrap_marked_paras_dc() early in the initialization
process.

This is not a problem for windowed Rich Edit controls, which already
calls ME_WrapMarkedParagraphs() before the user or application starts
interacting with it.