Improve Benchmark Reporting

[ptheywood]

The benchmark reporting offers a mix of minimum repetition times, mean repetition times, with and without seeds depending on the benchmark.

It would be more useful to:

• record all runtimes during benchmarking to disk, for post processing
• Post-process benchamrk times to find min, mean, deviation etc.
• Seed runs for more deterministic benchmarking, though as models initialisation is stochastic and initialisation RNG varies multiple seeds should still be benchmarked