Change time limit computation.

For a measured time of 0.49 second, a prover would only get 1 second to
replay the proof, i.e., half of the expected 1.98 seconds. The time limit
is now rounded to nearest to prevent this underestimation on fast proofs.