Add missing pixel format documentation

Fixes #1027.

docs/input.dox

@@ -346,8 +346,8 @@ If cursor creation fails, `NULL` will be returned, so it is necessary to check
 the return value.
 
 The image data is 32-bit, little-endian, non-premultiplied RGBA, i.e. eight bits
-per channel.  The pixels are arranged canonically as sequential rows, starting
-from the top-left corner.
+per channel with the red channel first.  The pixels are arranged canonically as
+sequential rows, starting from the top-left corner.
 
 
 @subsubsection cursor_standard Standard cursor creation

docs/window.dox

@@ -785,6 +785,10 @@ images[1] = load_icon("my_icon_small.png");
 glfwSetWindowIcon(window, 2, images);
 @endcode
 
+The image data is 32-bit, little-endian, non-premultiplied RGBA, i.e. eight bits
+per channel with the red channel first.  The pixels are arranged canonically as
+sequential rows, starting from the top-left corner.
+
 To revert to the default window icon, pass in an empty image array.
 
 @code

include/GLFW/glfw3.h

@@ -1458,6 +1458,9 @@ typedef struct GLFWgammaramp
 } GLFWgammaramp;
 
 /*! @brief Image data.
+ *
+ *  This describes a single 2D image.  See the documentation for each related
+ *  function what the expected pixel format is.
  *
  *  @sa @ref cursor_custom
  *  @sa @ref window_icon
@@ -2323,6 +2326,10 @@ GLFWAPI void glfwSetWindowTitle(GLFWwindow* window, const char* title);
  *  selected.  If no images are specified, the window reverts to its default
  *  icon.
  *
+ *  The pixels are 32-bit, little-endian, non-premultiplied RGBA, i.e. eight
+ *  bits per channel with the red channel first.  They are arranged canonically
+ *  as packed sequential rows, starting from the top-left corner.
+ *
  *  The desired image sizes varies depending on platform and system settings.
  *  The selected images will be rescaled as needed.  Good sizes include 16x16,
  *  32x32 and 48x48.
@@ -3726,8 +3733,8 @@ GLFWAPI void glfwSetCursorPos(GLFWwindow* window, double xpos, double ypos);
  *  Any remaining cursors are destroyed by @ref glfwTerminate.
  *
  *  The pixels are 32-bit, little-endian, non-premultiplied RGBA, i.e. eight
- *  bits per channel.  They are arranged canonically as packed sequential rows,
- *  starting from the top-left corner.
+ *  bits per channel with the red channel first.  They are arranged canonically
+ *  as packed sequential rows, starting from the top-left corner.
  *
  *  The cursor hotspot is specified in pixels, relative to the upper-left corner
  *  of the cursor image.  Like all other coordinate systems in GLFW, the X-axis