SpikeMonitor should support recording sets of indices

[thesamovar]

See https://brian.discourse.group/t/how-to-record-only-certain-neurons-from-a-group-by-spikemonitor/167/5

I think we avoided doing this in the past, but if we are going to implement #1229 then why not this too?

[mstimberg]

Agreed, if we do #1229, then this would just be some minor refactoring so that record indices and non-contiguous subgroups are handled in the same way. Which reminds me that there is a subtle issue about what indices to use in that case, but let's discuss this in #1229 .