trie: make hasher benchmark meaningful post-caches

[karalabe]

This PR is an aid for #15131

The trie hasher benchmark had two significant flaws that made it non-significant:

• The data inserted into it didn't look like a realistic Ethereum trie. Trie keys in Ethereum are always hashes or addresses (either way random data), which makes the trie more or less balanced. The existing benchmark just inserts 20K incremental ints, making the tree unbalanced and inappropriate for concurrency benchmarks.
• The trie has an internal cache, which among other, caches the result of hashing operations too. As such, the cost of hashing the trie N times will always be the cost of the first hash, since subsequent ops are NOOP.

This PR makes a few modifications to make the benchmark nicer:

• Creates a realistc random account trie.
• Uses the benchmark iteration count for the number of entries.
• Reports memory allocs too for later concurrency testing.