Spike replay

[halfflat]

Wishlist item:

Spike replay feature with applications for validation:

• Re-inject spikes from a previous Arbor (or other simulation) run.
• Suppress delivery of natively generated spikes.

Details to be fleshed out later — this is a placeholder ticket.

More justification:

The primary motivation is for validating models ported from other simulator engines, based on spike reporting. Small differences in spike times can lead to wildly different spike plots owing to chaotic effects, even if the results would otherwise be statistically similar.

By inputting the spikes from one simulation run, and suppressing the delivery of generated spikes, we can compare spike plots much more reasonably, and define useful error metrics over them.