USB target server: How to handle timeouts?

Currently, there are a couple places that sending can hang:

If you publish (or reply), and the endpoint is not enabled, we will currently wait unbounded
If the endpoint is enabled, but the host is not accepting messages, we will currently wait unbounded

Users could handle these cases by wrapping all calls to send in timeouts, but this is unsatisfying.

We could pull in embassy-time, but that's another dep we have to manage (and manage versions of).

now_or_never works okay-ish for "host is connected or not", but doesn't work well for sending messages to a full endpoint.

jamesmunns / postcard-rpc