[swarm] MultiHandler: Respect inbound timeouts and upgrade versions.

[romanb]

At the moment the ProtocolsHandler::listen_protocol() implementation of the MultiHandler discards configured timeouts and upgrade protocol versions on the inner handlers. The former results in the MultiHandler always having a default 10 second timeout on inbound substream upgrades, regardless of the configuration on the inner handlers. The latter results in the MultiHandler always using the upgrade::V1 negotiation protocol for inbound substream upgrades, which may not match what the inner handlers use for their corresponding outbound substreams - of course it should still be the same version used in all handlers, but it may not be V1. So this PR does the following:

1. The MultiHandler uses the maximum of all configured inbound timeouts on the inner handlers. Typically these are all the same but need not necessarily be.
2. The MultiHandler uses the upgrade::Version used by the inner handlers, unless they are not all the same, in which case a warning is logged and V1 used.