CD daemon: facilitate large CD formation convergence by adding jitter

[jgehrcke]

A major finding while looking into #723 was that

1. the rate limiter implementations we use for task reconciliation do not use jitter
2. natural system jitter can be very small; and lock-step behavior can actually result in unnecessary amount of conflicts

For more detail, see #723 (comment)

Conclusion: at least for the CD daemon's reconciliation loop we have to implement a simple jitter to spread/smear out the updates performed by individual nodes a little more evenly across time.

[jgehrcke]

Done via #732.