Make the expected number of tasks per cell a parameter.

Previously we had a max of 96, much smaller numbers will work depending on the job

See merge request !395

Replace calls to rand() by calls to rand_r() in the scheduler

Implements #302 using @d74ksy's changes. 

See merge request !405

Cache auto vec

Auto-vectorise populating the cache with hints to the compiler and `restrict` keyword.

See merge request !404

Correct the use of rshift to reduce the number of distance calculations.

Applies `rshift` (boundary conditions) correctly in `DOPAIR1` + `DOPAIR2` to reduce the number of spurious distance calculations.

See merge request !403

Avx512 fixes

* Adds a generic mask, `mask_t`, for each instruction set (AVX, AVX2, AVX512)
* Removes testPair and testPairPerturbed which are superseded by test27cells and test27cellsPerturbed
* Adds testActivePair that computes the density between a pair of cells in various configurations of active particles (face, edge, corner)
* Replaces all arithmetic vector operations to support AVX512 as there is no compiler support to overload them
* Computes max_index_i[] for each particle in runner_dopair1_density_vec to follow ParCo paper
* testInteractions now calls correct vectorised density particle interactions and is run as a part of the test suite
* Solves #327


See merge request !396

Updated tolerances because the command line argument, -p, is now used in test27cells.c to perturb the smoothing length and not just the default value of 1.1f.

Added extra test case to test27cells.sh and test27cellsPerturbed.sh to perturb the smoothing lengths further to pick up extra edge cases.

Fix scheduler steal policy

Set `scheduler` steal policy from `engine` policy. 

Solves #347.

See merge request !399

Conflicts:
	tests/testPair.c

Yamling

Warn when overriding parameters from the command-line.

See merge request !402

No need to explicitly activate the drift_part task in engine_marktask()
All the required drifts will be activated when we loop over the density pairs and call `cell_activate_drift_part()`. 

See merge request !401

Make dumpCells function dump all locally active cells and perform a careful task count when SWIFT_DEBUG_CHECKS is defined

Hopefully still quite conservative