bugfix for consistency check failure when undoing a deleted [draw] command that has a scalar on the same canvas

revert a change to text_setto from 91cf2a84 that caused a subtle regression when a broken object gets retexted.  (It caused the object to be put at the end of the glist, which caused the consequent canvas_restoreconnections call to make a wire connection to the wrong object)

That whole function is a nasty mess that needs to be refactored

First stab at making the "Cancel" button in gatom dialog actually revert to the previously saved values.

only put a new message box in floating mode (i.e., following the mouse) if it doesn't have an autopatch connection to the object above it

changed the interface for canvas_objtext, but it's not public so that shouldn't be a problem

changed the interface for canvas_objtext, but it's not public so that shouldn't be a problem

*added message object's ability to parse $0 (as per http://lists.puredata.info/pipermail/pd-list/2014-09/107785.html)

fix regression with [pd] not respecting new box dimensions when renaming to [pd foo]. Also, comment out more sys_vgui calls in g_canvas.c

from pd-l2ork: 7ad47e188a941508b92e3c96bc9e05432fbb88f7

got a very early working version that does _all_ object text editing (arrow keys, selection, etc.) in the GUI. (However, it still depends on g_rtext.c for doing everything else.)