Add `cache` operator about reactor-core

[smaldini]

Some context:

It's only Mono we're talking about, cache is somewhat complicated due to the implication it turns lazily a source sequence into a shared state replayable sequence.

In Stream we use Stream#multicast(EmitterProcessor.replay(history)).autoConnect(). The whole point of multicast + autoconnect is to share a reference processor and coordinate the subscribe so it happens safely once upstream on first subscribe downstream.

For Mono, due to the nature of its volume things can be much simpler. Actually that's the point of Promise, caching and sharing one deferred scalar value, to 1 or N current/late subscribers.

[smaldini]

There are also 2 cache points we can use:

• Cache on receiving side as in Stream#cache that will preserve current and late subscribers from requesting or subscribing the decorated source.
• Cache on producing side, a sort of cached Mono#defer that will only preserve late subscribers from subscribing the decorated source.

It seems WebSession use case is more about the latter wdyt @rstoyanchev ?

[rstoyanchev]

Yes a cache on Mono would suffice for me.