Test Details Overview

Count closed connections toward the per IP limit for a cooldown (#5261)

* Count closed connections toward the per IP limit for a cooldown

A connection that closed keeps counting toward the per IP and per
subnetwork limits of `tcp_listener` for `tcp.reconnect_cooldown`, 60
seconds by default, so a peer gets `max_peers_per_ip` connections per
cooldown rather than per purge tick. Cooldowns do not count toward the
inbound connection cap.

* Comment the `tcp_config` settings

* Index listener connections by address and subnetwork

`tcp_listener` keeps its connections in a multi_index hashed by the
normalized address and the subnetwork, both stored on the entry, so
the per IP and per subnetwork counts are index lookups like the
cooldowns instead of scans over the list.

* Move the listener's limit bookkeeping into `tcp_listener_index`

`tcp_listener_index` is the listener's one table: the open connections
with their sockets and servers, the outbound connects in flight with
their tasks, and the cooldowns of closed connections, all counted
against the caps they are subject to. It answers whether one more
connection or attempt fits, hands removed entries back so the listener
closes them outside its lock, and never touches the handles it carries,
so the new suite drives it directly with injected time and null handles.

`connection_type` moves to the transport namespace beside `socket_type`
and `socket_endpoint`, the enum helpers become free functions, and
`common.cpp` takes the `to_string` bodies that sat in `tcp_socket.cpp`.

A full inbound or outbound cap now counts under its own stat detail
instead of `max_attempts`, and the global attempt cap refuses at the
cap like every other one.

__________

Type: commit

Hash: e8d0db009b9c21b7608c86623cbc80178f8e1193

Created: today

__________

gr0v1ty/nano-node:e8d0db009b9c21b7608c86623cbc80178f8e1193