The restart strategy on the slide server was such that it would _never_ stop in the case that the ets table it was trimming went away. Instead it would metronimically add a crash message to your log every 30 seconds. This commit sets the restart strategy to 3 failures in 3 minutes. Since the only seen live examples of a slide server crash have been when the ets table has gone away, it is best crash and stay dead than be restarted after 3 failed trims.
A crash of `folsom_metrics_histogram_ets` leaves spirals broken. The spiral ets table goes away, so it can not be updated, nor can it be deleted and re-created due to the delete order of metric table, spiral table, folsom table. Have changed the order to be more like that of `histogram` so deletes will succeed in the event of a missing `spiral` ets table.
…r-slide * Avoid EUnit-enforced timeouts with prop_window_test_() * Change folsom_sample_slide_test.erl to avoid failure like https://gist.github.com/272dc6bcd2478484bffa * Fix sorting bug Russell sees but I don't, sorry!