... Rather, construct them during hit tests (also capturing more state sooner
rather than at Click time, and adding some accessors for later use)
This also fixes bug 1677 by other means and avoids similar problems.
A cell may be implemented to re-use a previously hit handle object, not yet
clicked, in a later hit test, by remembering a weak pointer, but TrackPanel
holds the strong pointers that determine when the object is destroyed.
And the objects will surely be destroyed after drag-release, or ESC key.
For now they are also destroyed whenever not dragging, and hit-testing is
re-invoked; that will be changed later, so that the re-use mentioned above
becomes effective, but still they will be destroyed when the pointer moves
from one cell to another.
... Let cell hit tests, and handle preview, know states only, not transitions.
Cell hit tests are passed a mouse state that does not always match the current,
but anticipates the button click to come; usually left, but if the Control
[sic] key on Mac is down, then right.
Thus, pressing and releasing Mac Control in multi-tool switches in and out of
the magnifier cursor.
... those handles that force a simulated button-up when play interrupts drag,
and can assume that pointers to tracks remain nonNULL and part of the current
project state (not the undo history).
Also pass shared_ptr to Track into more hit test routines.
