Performance of (re-)generating large SGraphs

[mlange05]

With the new Scheduler (PR #213) approaching completion, one of the remaining issues flagged was the performance of building and re-building large SGraph objects, eg. for EC-physics. As a result we reverted to caching the SGraph and only rebuilding it when needed (after discovery and parsing). This issue is to track the necessary performance improvements needed to re-enable fast on-the-fly re-builds.

As a note, a quick initial investigation showed that some of the parser-class caching in make_complete is not quite working as intended yet, but a "quick fix" had severe knock-on effects. Moreover, even with this, there is probably still a significant amount of time spend using FindNodes to (re-)define the dependencies for each SGraph item, causing re-builds to do many traversals per rebuild, which could be improved.