Clean up code and documentation for glfwGetKeyScancode and add
implementation for Wayland and Mir.

Related to #830.

Allows retrieval of platform scancode from GLFW_KEY_*.  Implemented for
Win32, Cocoa and X11.  Stubs for Mir and Wayland.

Closes #830.

Related to #478.

There is a gap between the creation of the context and the call to
_glfwRefreshContextAttribs.

Due to Wayland, shared code cannot rely on cursor positioning being
supported by the underlying platform.

This implicitly fixes #617 as it moves cursor centering into
_glfwPlatformSetCursorMode, thus separating it from the stale value of
_glfw.cursorWindow.

Fixes #617.

Fixes #145.

This change does not affect the ABI.

This adds the ability to switch between windowed and full screen modes,
move a full screen window between monitors and update its desired
resolution and refresh rate.

Fixes #43.

Adds support for setting window icons programmatically on platforms
where this makes sense.

Fixes #453.
Closes #467.

This function will put the calling thread to sleep until an event
arrives or until the specified timeout has elapsed.

This removes the (undocumented) behavior where glfwShowWindow would
bring the window to front and set input focus.  That function now
does what it says.

Regression caused by 71c72db1.

Since the Wayland SHM buffer format is implicitly premultiplied and the
GLFWimage pixels are defined to be non-premultiplied, we need to
convert the non-premultiplied pixels to premultiplied when filling the
buffer.

Related to #707.

This implements support for the 'DISABLED' cursor mode, which
effectively means locking the pointer to the surface. The cursor is also
explicitly hidden.

This adds two new build dependencies: wayland-scanner and
wayland-protocols.

Closes #708.

Fixes #714.

Fixes #266.

Added GLFW_INCLUDE_VULKAN.  Added glfwVulkanSupported,
glfwGetRequiredInstanceExtensions, glfwGetInstanceProcAddress,
glfwGetPhysicalDevicePresentationSupport and glfwCreateWindowSurface.
Added port of LunarG SDK tri example.

Although very unlikely, the wl_compositor version might not support
wl_surface.set_buffer_scale while the wl_output emits a wl_output.scale
that is larger than 1. So for correctness, bail on changing the buffer
scale if we won't be able to set it later.

Windows now keep track of the monitors they are on, so we can calculate
the best scaling factor for them, by using the maximum of each of the
monitors.

The compositor scales down the buffer automatically when it is on a
lower density monitor, instead of the previous way where it was scaling
up the buffer on higher density monitors, which makes the application
look much better on those ones.

Also merge win32_tls.h into win32_platform.h.

Fixes #645.

Fixes #464.

This fixes a regression introduced in 496f559c where a context would be
created for a NULL egl_surface, which subsequently fails.

Fixes #633.

Closes #620.