cacheSolve() through to the solve() call.  Of course, there are two
problems:

1. setinverse() should probably check that the arg is NULL or a matrix of
the correct dimensions.  Right now, doing a x$setinverse(1) breaks things.

2. If cacheSolve() is called multiple times with different values of b,
and/or tol, then the cached inverse may be incorrect.  In a more
robust solution, both of these args would need to be cached and checked
for changes, as well.