These look like they should be passing in error codes, at the least boost::asio::error::fault

I thought about passing in error codes, but passing in asio::fault seems wrong as this is not an asio fault. The appropriate thing to do would be to create a nano::error_code::deserialization_error but I didn't think it was worth the effort. Will investigate how to do that.

To match other code an error code is signalled by returning true, success == 0 == false.

I found that convention very misleading, seems like a low level C legacy. Let's say we have a check for success using the false for indicating that: 'if !operation_successful())' that reads "if not operation successful" which is logically the opposite of what we want. Needless cognitive overhead in my opinion.

For consistency this should be renamed to handshake rather than undefined since it's referenced by is_handshake_connection.

But calling it handshake would be incorrect for bootstrap connections. Wouldn't it?
We could possibly have an additional state calling handshake when we have started a handshake but before completing it.

I think is_handshake_connection() should be renamed to is_undefined_connection() because that is exactly what it is checking.

Could be renamed, it's true that there won't be any handshake if the connection is used for bootstrapping. I would move the whole socket type tracking into bootstrap_server (and rename that to network_server) because at the end of the day socket should be oblivious as to what it's used for, this is the task for higher layers