Switch to async synchronization primitives

[dbcfd]

PR for #12

[ebkalderon]

Looks like this is mostly done in master by c7ef48c, but it seems that it was missed in NatsClientInner::on_connect (I'm ignoring the std::sync::Mutex usage in src/nuid.rs because it doesn't look heavily IO-bound). CC @awesomike

I also wonder why std::sync::RwLock was replaced with futures::sync::Mutex instead of tokio::sync::RwLock? 🤔 This library seems to already be heavily invested in tokio, with heavy structural reliance on tokio::spawn and tokio::io::{AsyncRead, AsyncWrite}, so why not go all-in? It's not a huge deal, though. The most important thing is avoiding stalling the executor with blocking calls; the choice between an async-aware Mutex or RwLock is arguably less important.