Safety: Sanity-cap the init-channel counts received from the opposing peer at session open. #36

Sanity-cap the init-channel counts received from the opposing peer at session open, internally via the SMC (session master channel) (server side: LogInReq.numInitChannelsByCliReq; client side: LogInRsp.numInitChannelsBySrvReq plus the local count). Today a nonsensical value from a buggy (or foreign... but again, we assume trust of the peer) peer can throw inside the worker thread's reserve() and terminate the process, or can wrap the size_t sum. Choose a cap and an error code, and document the limit on the advanced sync_connect() and async_accept() overloads.

P.S. This is a safety thing; really there's quite a bit of protocol over the SMC -- and some other data-channels/layers -- and weird/bad/huge values in various places could probably had bad effects. We assume trust, so such things are best-effort in any case.