Print to stderr after  flushing stdout

Closes #413

See merge request !535

Speed up redistribute

See merge request !513

This reverts merge request !532

Add master option for subgrid model

Closes #422

See merge request !532

Allow the user to specify a non-standard path for the FFTW library at configure time.

See merge request !534

Escape the self-task recursion in the gravity sub-cell unskipping only if the cell has *no* gparts not if it has *some* gparts.

Improve unskip by escaping early

See merge request !528

Parallel initialisation

See merge request !533

Add support for Grackle 3.0.

See merge request !499

Note the quanta are nodes, not parts, this is necessary as we need to know where the starts of the sorted indices are located

calloc is faster than malloc+memset

Also formatting, sorry should have commit before doing that