I've branched well off of upstream glloadgen since they don't appear to be very active in between GL versions.

1) Changed how the address finder functions are set to be more sane.
2) Added *all* possible extensions. For the lulz.
3) Fixed a retarded bug in IsVersionGEQ.
4) Slapped myself for not seeing IsVersionGEQ bug earlier. Most important change.

- Implement texture scaling/conversion/downloading for the main view so
  we can finally start getting data to output.

  Also, redesign how it works a bit, it will now properly wait one full
  frame for each step in the process:  rendering the main texture,
  scaling the main texture to an output texture, staging/downloading the
  ouput texture, and then outputting that staged data.  This way, the
  GPU will have more than enough time to fully complete each step.

- Fix a bug with OpenGL plugin's texture staging function.  Was using
  glBindBuffer instead of what should have been used:  glBindTexture.

- Change the naming scheme of the variables in default.effect.  It's now
  named with the idea of just "color matrix" in mind instead of "yuv
  matrix", and instead of DrawRGBToYUV, it's now just DrawMatrix.

The returned ref count was never used and the returned value doesn't
actually confer any meaning in a multithreaded environment

Move CPack win32 generator from NSIS to WIX

Doing this ensures Qt thread safety when adding/removing sources and
scenes, because it will defer the function call to the main thread.

Replace a NULL with nullptr and reduce code duplication with the =
operators

Also, removed some unused headers, and moved the constructor/destructor
functions to the top

- Implemented better C++ classes for handling scenes/sources/items in
  obs.hpp, allowing them to automatically increment and decrement the
  references of each, as well as assign them to QVariants.

- Because QVariants are now using the C++ classes, remove the pointer
  QVariant wrapper.

- Use the new C++ classes with the QVariant user data of list box items,
  both for the sake of thread safety and to ensure that the data
  referenced is not freed until removed.  NOTE: still might need some
  testing.

- Implemented a source-remove signal from libobs, and start using that
  signal instead of the source-destroy signal for signalling item
  removal.

- Add property list callbacks to sources/outputs/encoders so that if
  necessary user interface can be automatically generated or perhaps a
  property list widget can be used for them.

- Change some of the property API names.  obs_property_list_t felt a bit
  awkward when actually using it, so I just renamed it to
  obs_properties_t.

- Removed the getdata/setdata nad getparam/setparam functions from
  sources/services, they will be superseded by the dynamic procedure
  call API.

Implement a properties definition interface to allow modules to export
general properties associated with objects of libobs.

The properties definition interface allows the option for automatic
settings UI generation (which will make simple plugins easier to develop
without the need for user interface), as well as allow real-time
property editing of values of things like sources/outputs/etc without
having to open property dialogs.  More property types can be added in
the future as needed as well.

Reduce and simplify the UI export interface.  Having to export functions
with designated names was a bit silly for this case, it makes more sense
for inputs/outputs/etc because they have more functions associated with
them, but in this case the callback can be retrieved simply through the
enumeration exports.  Makes it a bit easier and a little less awkward
for this situation.

Also, changed the exports and names to be a bit more consistent,
labelling them both as either "modal" or "modeless", and changed the UI
function calls to obs_exec_ui and obs_create_ui to imply modal/modeless
functionality a bit more.

I realized that I had intended modeless UI to be usable by plugins, but
it had been pointed out to me that modeless really needs to return a
pointer/handle to the user interface object that was created.

The ui_enum function gets a const struct obs_ui_info **, which basically
means it expects static data to be used.  I originally had it the other
way around, but yea, it's probably not a good idea, so I'm going to
revert back to the original code instead, which doesn't rely on the data
being static.

Made it so that enum_ui returns a const pointer to a structure rather
than require an actual structure.

Changed a few of the descriptions that I missed.

Add the ability to be able to call and use toolkit-specific or
program-specific user interface in modules.

User interface code can be either bundled with the module, or 'split'
out in to separate libraries (recommended).

There are three reasons why splitting is recommended:

  1.) It allows plugins to be able to create custom user interface for
      each toolkit if desired.

  2.) Often, UI will be programmed in one language (the language of the
      toolkit), and core logic may be programmed in another.  This
      allows plugins to keep the languages separated if necessary.

  3.) It prevents direct linkage of UI toolkits libraries with core
      module logic.

Splitting is not required, though is recommended if you want your plugin
to be more flexible with other user interface toolkits or programs.

Will implement a generic properties lookup next, which will be used for
automatic UI handling so that plugin UI isn't necessarily required.

Improve ALL_BUILD.vcxproj.user copy check

git describe based version number support for CPack

Also don't code in anger

The was_removed variable was rather unnecessary

Scene items previously were removed by calling obs_sceneitem_destroy,
but this proved to be a potential race condition where two different
threads could try to destroy the same scene item at the same time.

Instead of doing that, reference counting is now used on scene items,
and an explicit obs_sceneitem_remove function is used instead for item
removal, which sets a 'removed' variable to ensure it can only be called
exactly one time.

The previous commit used the scene as a  parameter to check to see if
the scene item was still present within the scene before destroying, but
this was actually unnecessary because the fault was because the destroy
signal was being triggered *before* the scene's mutex locked, thus
causing a race condition.  I changed the code so that it signals after
the lock instead of before, so the scene parameter should no longer be
necessary.