My understanding (you'd have to get a response from Matt Venn to be certain) is that this is driven from a state machine, with separate signals for the scan clock and the projects' clock, so that the state machine can put an arbitrary amount of dead time between the end of the scan load and the clock driving the projects. If done with with the two clocks non-overlapping, then both setup and hold violations can always be avoided by slowing the system down sufficiently.