* For example, if the PHY linkup is down, instead of looping until resumption of the linkup, a write operation now closes the socket for re-connection in the future
* Based on quartiq's minimq as of 933687c2e4
* In minimq applications, a socket is expected to be returned when `nal::TcpStack::open()` is called
* `MqttClient::read()`/`write()` takes away the TCP socket handle (wrapped as an `Option`) from its `RefCell`, and then calls `nal::TcpStack::read()`/`write()`; if NAL returns `nb::Error`, then the MQTT client will propagate and return the error, leaving `None` behind
* Afterwards, when `MqttClient::socket_is_connected()` gets called (e.g. while polling the interface), it will detect that the socket handle is `None`, and attempt to call `nal::TcpStack::open()`
* Since `open()` pops a socket from the array (`unused_handles`), when implementing this NAL the socket should have been pushed back to the stack, i.e. by `close()`; this prevents any future calls of `open()` from returning `NetworkError::NoSocket` due to emptiness of the array of socket handles
* `NetworkStack::connect()`:
* Add timeout for connection attempt
* Now returns the socket at TCP ESTABLISHED or CLOSED states, or after connection timeout
* Split `NetworkStack::update()` into `update()` (for controlling the clock) and `poll()` (for polling the smoltcp EthernetInterface)
* Also remove option `auto_time_update`; the main application is responsible for what values `embedded_time::clock::Clock::try_now()` should return