Add throttling to node prefetchers

[HerbertJordan]

This PR introduces two improvements to the parallel visit code:

• throttling of node pre-fetchers: workers stop prefetching nodes beyond a configurable threshold of prefetched nodes; however, since at the time workers are throttled not all needed nodes may have been fetched, the main thread is extended to contribute to the prefetching as needed, this puts a constant upper limit on the memory usage of the visitAll operation;

• lockstep prefetching of workers: workers will synchronize their activities to avoid prefetching large shares of far-future nodes and thereby exceeding the throttling limit for a long time. By synchronizing prefetch workers to load nodes in blocks, the focus of what is to be prefetched is forced to remain on the near future window.

For the latter, a barrier utility was introduced facilitating the synchronization of workers.

• complete a full export run for block 88M

[kjezek]

nice idea of how to enable the throttling, thanks