This is just enough for loadPixel/savePixel, but if other methods
release the GVL, potentially all other callback functions could
be called. These are now yet wrapped by gvl-aware wrappers.

without check.

This solves crashes of kind:
	[BUG] rb_gc_mark(): 0x00000003f54af8 is T_ZOMBIE

All parameters to Ruby event blocks or overloaded methods are converted to
proper Ruby classes or wrapped by appropriate wrapper classes for borrowed
objects. The wrapper classes are registered, to keep the connection between
C-pointer (the borrowed object) and the wrapping Ruby object. This connection
is canceled either when the borrowed object is destroyed or the Ruby wrapper
is free'd. If neither of them happens, the wrapper object is re-used for the
next call.

This worked good with Ruby-1.8, which has one big mark and sweep run.
Ruby-2.2 introduced incremental GC and has several mark and sweep runs, so
that marks and sweeps are quite independent. Therefor we re-used objects,
that are already maked for sweeping (so called T_ZOMBIE).

This patch releases the connection between borrowed object and it's wrapper
object after each call into Ruby code. Therefore the wrapper object isn't
re-used, but a new one is created for each param to each call. The wrapper
objects are later free'd by the GC. This patch could have a negative
performance impact.

Static linking all dependent libraries into each ruby version (from
1.8 to 2.2) would blow up the gem to >20MB.

This way it also works for derived classes like FXJPGImage.

Allow retrival of parts of the pixel data.

../../../../ext/fox16_c/FXRbApp.cpp: In static member function ‘static FXRbApp* FXRbApp::constructAndInit(const FXchar*, const FXchar*)’:
../../../../ext/fox16_c/FXRbApp.cpp:58:28: warning: deprecated conversion from string constant to ‘char*’ [-Wwrite-strings]
   static char* argv[]={"",0};
                            ^

This fixes issue #10.

Stacktrace was:
==4561== Invalid read of size 8
==4561==    at 0x4F726FF: st_lookup (st.c:414)
==4561==    by 0x4FCEAD6: rb_method_entry_get_without_cache (vm_method.c:182)
==4561==    by 0x4FD94E5: rb_funcall (vm_eval.c:315)
==4561==    by 0x810DAF2: FXRbCallVoidMethod(FX::FXObject*, unsigned long) (in /home/lars/.rvm/gems/ruby-2.0.0-p247/gems/fxruby-1.6.28/lib/fox16_c.so)
==4561==    by 0x81AC98E: FXRbTreeList::recalc() (in /home/lars/.rvm/gems/ruby-2.0.0-p247/gems/fxruby-1.6.28/lib/fox16_c.so)
==4561==    by 0x8FC78DC: FX::FXTreeList::removeItems(FX::FXTreeItem*, FX::FXTreeItem*, unsigned char) (FXTreeList.cpp:2340)
==4561==    by 0x8FC7916: FX::FXTreeList::removeItem(FX::FXTreeItem*, unsigned char) (FXTreeList.cpp:2347)
==4561==    by 0x829ED0F: FXRbTreeItem::freefunc(FX::FXTreeItem*) (in /home/lars/.rvm/gems/ruby-2.0.0-p247/gems/fxruby-1.6.28/lib/fox16_c.so)
==4561==    by 0x4EBA505: finalize_list (gc.c:1415)
==4561==    by 0x4EC15A3: rb_gc_call_finalizer_at_exit (gc.c:1552)
==4561==    by 0x4EA8064: ruby_cleanup (eval.c:127)
==4561==    by 0x4EA8292: ruby_run_node (eval.c:307)
==4561==  Address 0x74fb894853ff8550 is not stack'd, malloc'd or (recently) free'd
==4561==
segfault_on_close.rb: [BUG] Segmentation fault

So we need to use it properly too.

See bug https://bugs.ruby-lang.org/issues/6504#change-38019

Update cross compilation tasks to build x64 and x86 targets.

Before commit af543822 loading order of fox16.so and fox16.rb was irrelevant.
But now fox16.rb must be called, in order to load ruby files to the C++
extension. Renaming the C++ extension ensures, that fox16.rb is called first.

The internal object allocated when calling FXSplashWindow.new was FXSplashWindow instead
of FXRbSplashWindow, so that no FXRbUnregisterRubyObj() was called. Therefore the object
was double free'd.

Stacktrace was:
==11529== Invalid read of size 8
==11529==    at 0x9265BBC: FX::FXObject::getClassName() const (FXObject.cpp:280)
==11529==    by 0x862EE55: FXRbObject::freefunc(FX::FXObject*) (markfuncs.cpp:77)
==11529==    by 0x4EA6254: finalize_list (gc.c:2944)
==11529==    by 0x4EAAFC2: rb_gc_call_finalizer_at_exit (gc.c:3075)
==11529==    by 0x4E9773B: ruby_cleanup (eval.c:101)
==11529==    by 0x4E979C4: ruby_run_node (eval.c:244)
==11529==    by 0x4007FA: main (main.c:38)
==11529==  Address 0x75e24c0 is 0 bytes inside a block of size 248 free'd
==11529==    at 0x4C2A4BC: operator delete(void*) (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==11529==    by 0x92B98AD: FX::FXSplashWindow::~FXSplashWindow() (FXSplashWindow.cpp:219)
==11529==    by 0x917C990: FX::FXComposite::~FXComposite() (FXComposite.cpp:304)
==11529==    by 0x929A5EF: FX::FXRootWindow::~FXRootWindow() (FXRootWindow.cpp:209)
==11529==    by 0x929A621: FX::FXRootWindow::~FXRootWindow() (FXRootWindow.cpp:211)
==11529==    by 0x9156DB9: FX::FXApp::~FXApp() (FXApp.cpp:4994)
==11529==    by 0x84912F3: FXRbApp::~FXRbApp() (FXRbApp.cpp:118)
==11529==    by 0x8491347: FXRbApp::~FXRbApp() (FXRbApp.cpp:127)
==11529==    by 0x862EF8E: FXRbObject::freefunc(FX::FXObject*) (markfuncs.cpp:92)
==11529==    by 0x4EA6254: finalize_list (gc.c:2944)
==11529==    by 0x4EAAFC2: rb_gc_call_finalizer_at_exit (gc.c:3075)
==11529==    by 0x4E9773B: ruby_cleanup (eval.c:101)

Don't include sys/time.h for cross compilation, this avoids the fix for incompatible types of gettimeofday()

Allow more consistent read/write access to image data, that is also
usable with derived classes like FXJPGImage.

Moreover much faster String based access to the pixel data is possible,
now.

FXImage#data and FXMemoryBuffer is now deprecated.

Calling other classes while garbage collection is generally dangerous and useless.

This happens with FXFileSelector.

FXWindow destructor calls recalc() and changeFocus() of it's parent windows.
Since these methods are routed back to Ruby code, but calling Ruby code from
GC isn't a good idea, we mark the parent window as "in_gc", so that it will
ignore recalc() and changeFocus() calls completely.

The parent window should also be scheduled to be free'd. In the other case,
the child window would have been marked as used.

Calling any Ruby code within the mark phase is generelly dangerous,
because it usually allocates some objects.
Since most of the methods of FX objects can be overloaded in Ruby,
the calls in markfuncs now call explizitly the C++ methods.

Also remove some double marking of objects.

The target objects stored in the accelerator table are currently
private. Therefore FXRbAccelTable::markfunc() doesn't mark them as used.
As a workaround the objects are additionally stored in @targets Hash.

fix retrieving of file handles, so that FXApp#addInput works with ruby 1.8.7, 1.9.2 and rubinius under Linux (win32 not tested yet)