Refactor to allow reconnection to SHiP endpoints after connection lost.

[yarkinwho]

Resolve eosnetworkfoundation/eos-evm-contract#579

The changes:
1 Connection is moved from plugin_init to plugin_startup
2 Most network related errors will result in reset_connection() instead of sys::error()
3 Some errors such as cannot read block from db will still cause sys::error()
4 reset_connection() will send a close frame to existing connection (regardless success or not) and try resetting the connection.
a it will delay a configurable time before trying reconnection (default 10s)
b it will try at most a configurable times (default 0, means it will not try reconnection at all), after reaching the limit it will call sys::error() to exit
c If the system successfully reconnect to SHiP and read blocks properly, the reconnection count will be reset to 0

The PR also include a minor fix to work around the issue:
eosio::json_token_stream will fail if the input string is not zero terminated.
Fixed by adding the terminating zero.

[taokayan]

looks good to me

[arhag]

There were two issues this change exposed.

1. One was fixed by HashState Stage: skip check_block_sequence in unwind phase silkworm#38.
2. The other is tracked by Got "header to verify not found" error if the chain is unwinded and replayed from earlier blocks silkworm#39.

[yarkinwho]

For eosnetworkfoundation/silkworm#39

It is decided that we should fix it by not calling insert and verify for blocks that are already in canonical chain. Need update this PR.

[yarkinwho]

Updated the logic.

Now incoming duplicated eos blocks will be discarded instead of triggering the fork logic.