It is now easy to access any attribute by piping the output of
`query -T` to a JSON extractor/filter.
E.g.:
bspc query -T -d DESKTOP_SEL | jq -r .layout
And it also makes `restore -T` more robust.
Stacking now involves 3 layers: BELOW, NORMAL and ABOVE.
In each layers, floating windows are stacked above tiled windows.
The *stack* function is now extremely simple: it just inserts an item in
a sorted list, relying on *stack_cmp* to compare clients.
Fullscreen windows are no longer special.
This is a recurring request (the probability of removing an important
file is probably low).
Also: don't start the X setup unless we have a working socket.
Because many clients (e.g. termite) prevent us (maybe unknowingly) from
capturing motion events on their windows, we're forced to create a
window for this sole purpose.
Grabbing the pointer isn't an option, because it forces us to consider
some of the enter notify events we should be ignoring.
It appears that unmapping windows might generate undesirable enter
notify events. It can lead a very nasty bug where the focus constantly
switches between two desktops ad infinitum (cf. #268).
Tags should generalize desktops.
To accomplish this, the main node attributes: (type, ratio) would have
to become a dictionary: ((tf1, (type1, ratio1)), (tf2, (type2, ratio2),
...). (`tf<n>` being a tag field.).
The last desktop and monitor are now deduced from the history.
The stacking order is now independent from the history of the focused
nodes: this prevents hacks on both sides.
All windows are now considered in the stacking algorithm: it prevents
tiled windows from one monitor to appear above the floating windows of
another monitor.
Transfered windows are stacked below the windows of the same kind.
