cmd/evm: allow state dump regardless if test passes in statetest

[sorpaas]

This just makes it simpler for other evm implementations testing against geth.

[holiman]

In theory I agree, and I've been on the verge of making this PR myself, but when I've actually experimented with it, I've noticed that the state dump after a statetest cannot be relied upon. I started digging in that, but then got busy with other tasks.

As I recall it,

• The preimages aren't collected during statetest execution, so the dump misses addresses,
• There's something about flushing that is wrong, so when we iterate the trie we aren't actually getting everything.

The postCheck method get's a new state:

	statedb, _ = state.New(root, statedb.Database(), snaps)

And is invoked with the fresh 'state' but reused snaps`.

		postCheck(result, snaps, statedb)

However, before we constructed this new state, we did a Commit operation and discarded the nodes, so we never actually stored them to db.

	// Commit state mutations into database.
	root, _ := statedb.Commit(block.NumberU64(), config.IsEIP158(block.Number()))

A good start here would be to add some testcases that shows the failure of the current mechanism. Having proper tests here would be great (and making it work would probably not be too hard once we have tests)

[holiman]

Hm, I might be misremembering. When I was investigating this: ethereum/execution-spec-tests#332 , I noticed the shortcomings. However, that was regarding the blocktest command, not the statetest command.

[holiman]

LGTM