Adding collective modeling functions in model-net (currently only dragonfly/torus models support it)

- NOTE: shim just forces packet size to a huge number

- NOTE: the current scheduling policy of FCFS is hard-coded, this will change
  in future revisions
- NOTE: random number generation has changed due to the differences in
  scheduling and packet ordering. Hence, results may be slightly different from
  runs prior to this commit

- currently stupid, stats don't really mean
  anything when pulling
- enable manually by setting do_pull = 1 (no
  cmdline option yet)

- pull itself not yet tested

Instead of specifying the loggp config file path relative to the
path of the respective executable file, you now must specify it
relative to the path of the master config file.

This prevents simulations that run for a certain duration from
crashing rather than ending when simulation time is reached.

minor changes to model-net plug-in: (1) making the torus network's chunk size configurable (2) handling some edge cases which we haven't encountered so far but its good to have them for future

- these calls should be made by the top level model only, so that all
  results are consolidated into the same subdirectory
- add corresponding calls to modelnet test programs

- save previous cumulative time values rather than actually reversing
  calculation in order to avoid minor discrepancies introduced by
  rounding error