The `birth_mode` was not sufficient for properly unrotate the brother of
a removed node. The `birth_rotation` stores the rotation applied to the
brother subtree when the node was inserted.
This should help to enforce that (L + n) - n = L in every possible
cases. Which means that adding a node and removing it should leave the
layout unchanged.
The following 'grab_pointer' arguments are removed: 'move_tiled',
'resize_tiled'. The regular 'move' and 'resize_*' arguments shall be
used instead. The 'fence_grip' parameter is no longer meaningful and
has been removed. Moving and resizing now behaves the same for tiled and
floating windows.
'send_to' was broken, '--next' and '--prev' can't be options, and I
don't want to restrain the name of the desktops with 'send_to
DESKTOP_NAME|next|prev [--follow]' or handle multiple options: 'send_to
(DESKTOP_NAME|(next|prev --relative)) [--follow]' so I just created new
messages.
This teaches the send_to_monitor command to send a client to either the
next/prev monitor rather than having to always specify a named monitor.
Such options are useful to be able to send clients to monitors relative to
their own position.
It will facilitate debugging and also provide an exhaustive
representation of the current state that might be used to reconstruct
the layout of each desktops. While doing so, it occurred to me that the
*born_as* attribute of the *node_t* structure was only used for leaves,
so I changed it into a *client_t* attribute.
Two options have been added to the "cycle" message: --skip-class-equal
and --skip-class-differ. The class name of the client is now shown in
the output of the "dump" command instead of the letter "C".
