change space_getcell to space_getcells, can pick up more than one cell at a time.

As discussed in #255. Have only checked that it works, haven't done any benchmarks, but this can only make things faster 😉 

See merge request !300

Call vector

Call vectorised version of `runner_doself1_density` when vectorisation is turned on.

See merge request !302

Particle caching

Summary
-------
- Implements a vectorised version of `runner_doself1_density()`. 
- Implements a particle cache on each thread and allocates the memory `in engine_init()`
- Modified `test27cells.c` so that particles that have been interacted with vectors are checked manually against a serial interaction
- Implemented another vectorised version of `kernel_deval()` which calculates the kernel for two particles.
- Fixes a bug in `runner_iact_vec_force` and `runner_iact_nonsym_vec_force` when calculating the SPH term

`test27cells` will fail because of the thresholds set for the diff, I have attached the differences in: [brute_force_h_1.2348.dat](/uploads/8e32fd88cee93bd429bc57c64ab16b4d/brute_force_h_1.2348.dat)
[swift_dopair_h_1.2348_avx_doself1_loop.dat](/uploads/4c61c74b5d4a7aef8a2c0d01cdc17e9e/swift_dopair_h_1.2348_avx_doself1_loop.dat)

@nnrw56 @pdraper



See merge request !287

Check that the particles interacting in the SELF or PAIR tasks have been properly drifted to the current time

Particle logger

Next step in implementing streaming IO: add a `logger` that reads/writes `parts`, `gparts`, and timestamps to a `dump` (following #126).

Comes with a nice unit test!

Note that this does not yet touch the rest of the code, so it should have no effect on the rest of SWIFT. The next step will involve implementing a global dump and a new task type that actually writes particles in need of a move.

See merge request !299