Refactor drm backend (v3)

[Drakulix]

Still work in progress (docs and anvil support as well as some new examples are missing).

Ready for a first review.

The biggest change is the split of the DrmDevice and DrmBackend into

• LegacyDrmDevice and LegacyDrmSurface
• GbmDevice and GbmSurface
• EglDevice and EglSurface

To be able to do this clean seperation new traits have been added.
Most importantly Device and Surface, describing drm devices in general

• RawDevice and RawSurface providing the lowest layer and modesetting support, while the higher-level devices rely on other means of swapping buffers.

• LegacyDrm* implements the Raw*-traits, Gbm* requires a Raw*-object as it provides buffer management, but itself does not provide Raw*-implementations anymore, but instead the more high-level page_flip, recreate and unlock_buffer functions.

• EglDevice can wrap any Device that also implements egl's NativeDisplay trait (and where it's surface also implements NativeSurface), which is provided by the GbmDevice (and GbmSurface), through the methods listed above.

This provides very high composability, as the usual (and currently only viable) chain of an egl-gbm-drm-device, is not dictated anymore by the implementation, but by public-accessible traits.

In future this will allow us to add the following implementations easily:

• AtomicDrmDevice to support atomic modesetting (once drm-rs current refactor is done)
• EglStreamDevice to support nVidia's buffer mangement
• VulkanDevice to support Vulkan instead of OpenGL as a rendering api
• ... (probably some more)

• this also allows users of the library to provide their own implementations for more exotic devices.

During the refactor some additional changes have been made:

• Usage of the calloop 0.4 update to utilize EventedFd instead of EventedRawFd
• Seperate egl and gl modules
• Provide a raw GL loader additionally to the mostly deprecated (although still working) glium wrapper
• Renaming the GraphicsBackend to the more fitting CursorBackend
• Don't let any graphical backend depend on CursorBackend anymore
• Rebuild the udev-backend, so that it no longer creates devices (move this responsibility to the library user). This greatly simplifies the backend (removing most of it's weird api quirks). Additionally this makes working with the new drm-devices much nicer, as the concrete type of the device is not dictated by the udev backend anymore.

[elinorbgr]

Can't there be some link between the backend and the returned Gles2 ? Maybe adding a PhantomData lifetime to Gles2 ?

[Drakulix]

Sadly not. This is auto-generated code from gl_generator and it would probably also cause a lot of inconveniences for the user. GL Contexts are weird and many gl functions are unsafe anyway. If you want a safe alternative there is glium.