Fix timeout, cancellation and instant display logic

Bug: 20668802
Bug: 20484451
Bug: 20734909
Bug: 20738335

Change the timeout implementation to not rely on removing scheduled
timeouts when an evaluation completes.  This seems simper, more clearly
correct and, unlike its predessessor seems to work reliably.
Correctly display "cancelled" message, but only when cancelled for a
reason other than timeout.

Change SyntaxError to not be a Java "Error", since we routinely recover
from it.

Better support evaluation of partial expressions by removing trailing
operators.  Make some attempt to avoid redundant evaluations, since
they're not as cheap as they used to be.  Change the display logic to
avoid frequent clearing of the display, which resulted in
visible flashing.

Note that the evaluation logic can now stop evaluating before the end
of an expression.

Substantially lengthen timeout used when computing an unrequested
result.  Inverse trig functions can be slow.  See b/20729963.

Make large factorial computations interruptable, as I mistakenly
thought they were.

Ignore "=" on an empty expression.

I'm not positive this fixes the b/20484451, but I haven't been able to
reproduce it with this patch.

I have no idea whether it addresses b/20738335, but it cleans up code
in that vicinity, so it would be interesting to know whether that is
still reproducible.

Change-Id: I67af5afc00e19a6156e1547ce4e2060c70ca367f