New solver using SIMD types from the portable_simd project. This
currently requires nightly to build. It is a port of the existing
OpenCL IPM solver to Rust using SIMD. It is written using f64x4
types which will require the AVX2 instruction set.

Included is a genral refactor of the OCL IPM solver to improve
feature gating and code sharing with the SIMD solver.

Some tweaks to benchmark code.

Initial commit of SolverFeature.

Update benchmarking code to only time the simulation, and not
the seutp of the state / solvers.

Main solver is generic to the floating point type and the number
of lanes. The implemented solver is generic only for the number
of lanes.

Revised benchmarks to test differing lane sizes.

- Use u8 for solve status (reduces memory required)
- Put iteration loop in the kernel.
- Use num_chunks instead of chunk_size in OCL settings.
- Update random models to only create data for the requested
number of time-steps.

…benchmarks.