In other words, if the result for a state is q, and the bound p falls in the range
[q-e,q+e] where is the error bound from model checking computations, then an
error is generated rather than producing a potentially incorrect true/false result.
Explicit model checking engine only for now.
Test cases are added, and some (unrelated) existing regression tests with dubiously
close bound checks are fixed.
Both types of engine now disallow negative/infinite/NaN rewards.
Symbolic engines will mostly still not detect the presence of NaN
due to the way that CUDD deals with this.
Regression tests also added.
Dave Parker