merged

option parallel_proofs_reuse_timing controls reuse of log information -- since it is not always beneficial for performance;

prefer outdegree in comparison again (cf. 88c96e836ed6) -- NB: big jobs might hide behind small ones in this naive queuing scheme;

more tight representation of command timing;
tuned signatures;
tuned;

prefer comparison of session timing, if this is known already;
some odd workarounds to prevent crashes due to TreeSet.- sometimes not deleting the element;

proper check of Proof.is_relevant (again, cf. c3e99efacb67 and df8fc0567a3d);

support nested Thy_Syntax.element;
more explicit Keyword.is_proof_body;
tuned;

back to non-schematic 'sublocale' and 'interpretation' (despite df8fc0567a3d) for more potential parallelism;

help JVM to cope with large symbolic structures;

improved scheduling of forked proofs, based on elapsed time estimate (from last run via session log file);