test: make shadow_tracking more reliable

[Longarithm]

Fixing issues:

• _get_final_block_height' result is not a final block and expectation of min_height + 1 >= max_height fails sometimes. Getting minimum height instead.
• We assumed that if we first query shard assignment and then block with chunks, then it will correspond to the same epoch, but it's not necessary the case. Instead, it is more reliable to just get block which everyone has and then retrieve epoch id and its assignment.
• Making number of nodes = number of shards so that if shadowed account is incorrect, the test fails very quickly.
• Also it won't hurt to ensure that all epoch ids in rounds should be unique.

[marcelo-gonzalez]

It passes on my laptop, but on my desktop it seems this has uncovered some bug :/ . In test1 logs (built from ab75966):

thread 'tokio-runtime-worker' panicked at core/store/src/trie/shard_tries.rs:441:9:
assertion failed: !self.0.mem_tries.read().unwrap().contains_key(shard_uid)

[staffik]

Thanks Alex!