dsm.cor should be smarter

[dill]

Following some chats with Lindesay and Monique I realised that dsm.cor is being pretty dumb when it comes to plotting the correlogram. It probably makes sense to allow some better subsetting so we aren't just getting the aggregated effects of many many transects on top of each other, disguising what autocorrelation might be there. Some kind of faceted/trellis graphics might make sense to visualise the correlogram.