Identify objects that are leaking in the exchange benchmark. Deconstruct the slog

[rowgraus]

No description provided.

[warner]

We think we fixed this, by a combination of XS bugfixes (which were retaining objects too long) and liveslots improvements (doing the same).

We need to look at @mhofman 's current loadgen/perftest data to see if the kernel object table is growing over time (given a non-accumulative workload). We can re-open this if that data suggests that the problem isn't really gone.