Using ImTextureID instead of void* in ImFontAtlas, not sure why I didn't do that earlier, make things more clear

ImFontConfig: Removed MergeGlyphCenterV in favor of a more multipurpose ImFontConfig::GlyphOffset. (Breaking change)

ImFontConfig: Added GlyphOffset to explicitely offset glyphs at font build time, useful for merged fonts. May remove MergeGlyphCenterV soon.

Scrollbar: Avoid rendering when sizes are negative to reduce glitches (not sure how this ever slipped through - perhaps because of WindowMinSize settings).

To maintain maximum frame rate we render to the last acquired swapchain image but present the last but one drawn image. This behavior is optional through conditional compilation macros.

Removed redundant VkResult plugged into VkPresentInfoKHR pResults attribute. This is only meaningful when we present directly to multiple swapchains. In that case we can an VkResult per swapchain.

The transition of the swapchain image(s) can happen implicitly in the renderpass. This approach has been stated to be more efficient than using an explicit barrier.
See "Vulkan Programming Guide", Chapter 7: "Graphics Pipelines", section "Renderpasses".

Examples: Vulkan: Disable IMGUI_VULKAN_DEBUG_REPORT by default. Update .bat files for newer VulkanSDK.

Additional layer, extension and the callback itself are used/created when IMGUI_VULKAN_DEBUG_REPORT is defined. The callback reports seven (potential!) errors which will be fixed with another pull request.

Merge branch 'g_FrameIndex_not_used_fix' of https://github.com/ParticlePeter/imgui into ParticlePeter-g_FrameIndex_not_used_fix

Examples: Vulkan: Small changes to be in line with specs and fix validation layer errors/warnings

See e.g.
http://clang-developers.42468.n3.nabble.com/The-p-conversion-and-cast-to-void-td4044226.html

ImDrawList::AddImage* renamed uv_ parameters to be more consistent. Fix AddImageQuad to use IM_COL32_A_MASK (#1009)

AFAIK there's no such enum as `GL_BLEND_SRC` or `GL_BLEND_DST` in OpenGL3 and OpenGL4.

Maybe they're left over from previous versions? They don't even seem to exist in OpenGL 2.1. In fact checking they were deprecated in OpenGL 1.5 (they exist in 1.3)

Pros: Faster. Avoid early crashes StackId underflow that are meant to be more gracefully caught by CheckStacksSize() (with more explicit assert). Cons: Excercise less code.