Sort out naming of output cell dumps to be based on the step, so they can be compared to other dumps

The time weights where overflowing int (1<<32), so use double precision, which has the nice side effects mentioned above. We now also attempt to balance the weights bweteen vertices and edges, since they can have very different units

Naming was getting messy, so adopt a more obvious scheme of vertex/edge descriptions

Use timebins that are shared with the interacted cell, so that cells on the edge that are only active because of their interactions are also weighted

Also separate the scaling of weights bewteen vertices and edges, these are differing units so may not be similarily scaled

Only skip tasks with zero costs, note previous method had a bug and no weights for task_type_sub_pair was given

Use the equation of state functions in a few more places in the gizmo code to be more robust to changes.

Intel compiler does not like bit-comparison on enumerated types. Use an integer for the step properties instead.

Support for OSX

Closes #364

See merge request !442

Add information to stdout and timestep.txt about what happened in the step

See merge request !455

Make loop bounds consistent.

See merge request !452

Add information to stdout and timestep.txt about what happened in the step (rebuild, repartition, snapshot, ...)

Large hdf5 reads

See merge request !453

When compiling with the iosthermal equation of state, read the constant thermal energy from the YAML file.