trie: remove redundant returns + use stacktrie where applicable

[holiman]

Since we changed how generation works, and how the snap sync works, we stopped using some of the functionality provided by the prover.
Notably, we no longer make use of the returned Trie, nor the Notary. This PR removes them as return arguments. It also makes use of the Stacktrie for the cases where we don't have any proofs, and the provided key/value pairs are expected to produce a correct state root.
The first commit adds a new benchmark for this case, which exhibit quite a boost by the use of Stacktrie instead of the regular one:

name                      old time/op    new time/op    delta
Prove-6                      733µs ±21%     698µs ±27%     ~     (p=0.690 n=5+5)
VerifyProof-6               14.5µs ±16%    15.7µs ±15%     ~     (p=0.421 n=5+5)
VerifyRangeProof10-6         253µs ±16%     266µs ±15%     ~     (p=0.548 n=5+5)
VerifyRangeProof100-6       1.00ms ±18%    0.90ms ±22%     ~     (p=0.310 n=5+5)
VerifyRangeProof1000-6      13.4ms ±12%    13.1ms ±11%     ~     (p=0.690 n=5+5)
VerifyRangeProof5000-6      64.4ms ±10%    59.5ms ±21%     ~     (p=0.222 n=5+5)
VerifyRangeNoProof10-6      1.80ms ±25%    0.54ms ± 9%  -69.73%  (p=0.008 n=5+5)
VerifyRangeNoProof500-6     6.93ms ±48%    2.32ms ±21%  -66.47%  (p=0.008 n=5+5)
VerifyRangeNoProof1000-6    13.9ms ± 9%     4.3ms ± 3%  -69.33%  (p=0.016 n=5+4)

name                      old alloc/op   new alloc/op   delta
Prove-6                      144kB ±13%     140kB ±10%     ~     (p=0.548 n=5+5)
VerifyProof-6               5.51kB ± 6%    5.25kB ± 6%     ~     (p=0.151 n=5+5)
VerifyRangeProof10-6        61.2kB ±14%    66.3kB ± 6%     ~     (p=0.222 n=5+5)
VerifyRangeProof100-6        315kB ± 0%     315kB ± 0%     ~     (p=0.548 n=5+5)
VerifyRangeProof1000-6      3.76MB ± 1%    3.79MB ± 1%     ~     (p=0.421 n=5+5)
VerifyRangeProof5000-6      19.1MB ± 0%    19.1MB ± 0%     ~     (p=0.841 n=5+5)
VerifyRangeNoProof10-6       468kB ± 6%     103kB ± 3%  -77.93%  (p=0.008 n=5+5)
VerifyRangeNoProof500-6     1.88MB ± 1%    0.40MB ± 1%  -78.87%  (p=0.008 n=5+5)
VerifyRangeNoProof1000-6    3.56MB ± 1%    0.75MB ± 1%  -78.81%  (p=0.008 n=5+5)

name                      old allocs/op  new allocs/op  delta
Prove-6                      1.98k ±15%     1.90k ±10%     ~     (p=0.548 n=5+5)
VerifyProof-6                  106 ± 7%       102 ± 4%     ~     (p=0.111 n=5+5)
VerifyRangeProof10-6           639 ± 9%       667 ± 3%     ~     (p=0.310 n=5+5)
VerifyRangeProof100-6        2.50k ± 0%     2.49k ± 0%     ~     (p=0.516 n=5+5)
VerifyRangeProof1000-6       30.9k ± 1%     31.0k ± 1%     ~     (p=0.421 n=5+5)
VerifyRangeProof5000-6        158k ± 0%      158k ± 1%     ~     (p=0.421 n=5+5)
VerifyRangeNoProof10-6       4.45k ± 2%     1.51k ± 2%  -66.16%  (p=0.008 n=5+5)
VerifyRangeNoProof500-6      16.6k ± 1%      5.1k ± 1%  -69.08%  (p=0.008 n=5+5)
VerifyRangeNoProof1000-6     31.4k ± 1%      9.5k ± 1%  -69.63%  (p=0.008 n=5+5)

[holiman]

which exhibit quite a boost by the use of Stacktrie instead of the regular one.

Just to clarify: this improvement will be noticeable mainly when we're doing snap sync, and receive a small-to-medium size storage trie.

[karalabe]

LGTM