Restacktrie with pruning

[gballet]

This includes the changes that I made yesterday to keep memory usage low.

[gballet]

These are the benchmarking results btw:

ench Generation 
goos: linux
goarch: amd64
pkg: github.com/ethereum/go-ethereum/core/state/snapshot
BenchmarkTrieGeneration/4K/standard-8         	      73	  16872518 ns/op	 7563910 B/op	   97844 allocs/op
BenchmarkTrieGeneration/4K/pruning-8          	      48	  24518136 ns/op	 8001162 B/op	   94880 allocs/op
BenchmarkTrieGeneration/4K/stack-8            	      55	  20868011 ns/op	 7157821 B/op	  177513 allocs/op
BenchmarkTrieGeneration/10K/standard-8        	      30	  42971496 ns/op	20723174 B/op	  250570 allocs/op
BenchmarkTrieGeneration/10K/pruning-8         	      16	  71834673 ns/op	19991784 B/op	  237108 allocs/op
BenchmarkTrieGeneration/10K/stack-8           	      24	  52136639 ns/op	17890084 B/op	  443703 allocs/op
PASS

Allocs and execution times can still be optimized, they are not as good at the standard version but they already beat the standard pruning code.