Modifying dsmcFields to allow negative rhoNMean values, only cheching for min(mag(rhoNMean)) < VSMALL

Initialising rhoN, rhoM and iDof fields to VSMALL instead of 0.0 to avoid floating point errors or the need to skip field calculation.  Rewritten initialisation function to fill on a cell-by-cell basis, filling into randomly distributed positions in volume weighted selected tetrahedrons (decomposed cells) - much faster.

specified to be cached to avoid unnecessary recalculation.

equivalent to the 'none' entry in fvSchemes and relaxation factors must be
provided for all fields which might require one.  If it is set to a value
between 0 and 1 this value is used as the default relaxation factor.

Added to the forces functionObject the ability to read a force density (fD) volVectorField directly rather than a p and U field.  To use, the following entries go in the controlDict directForceDensity true; fDName      fDMean;

Added a resetOnOutput switch to fieldAverage to allow the averaging to be only of the preceding writeInterval for transient cases.

Removed boundary setting in DsmcCloud field calculation, calling correctBoundaryConditions in solver instead.

Cartesian subCelling of collision cell implemented.  Need to check the impact on performance and results

Changed way of populating database for dsmcFieldCalc.  Added correctBoundaryConditions() to sigmaTcRMax on initialisation

See http://openfoam.cfd-online.com/cgi-bin/forum/show.cgi?126/11559

Created new functionObject for dsmcFields, renamed post processing utility dsmcFields to dsmcFieldsCalc and calling function object from there.

Added an if (cR < SMALL) guard to VHS sigmaTcR calculation.  Giving each processor a different random number generator seed, otherwise they follow the same sequence and are not independent.  Tnis is a practical problem because they were producing molecules with the same velocity in neighbouring cells on different procs which then collided and caused a floating point error because they had equal velocities.