Allow consumers to set flags on keys

[shahab96]

Currently consumers have no way to set flags when writing to memcached. We should allow them to do so, however this would require changes in ProtocolTrait which will end up being a breaking change for current consumers. On the upside, if we're making a breaking change we have the opportunity to bring in async as well, which would also resolve #20 (not by using tokio specifically but by supporting async operations. Consumers can choose what async runtime they want to use)

[aisk]

I think we can implement this without change ProtocolTrait. Currently the set method accept a ToMemcacheValue as the value to store to memcached. The ToMemcacheValue trait have a method named get_flags, which will be used as the flag to memcached.

And we only implmented some basic types, like i32, String as a ToMemcacheValue, so users can't change the flags easily, but they can implement their own ToMemcacheValue to override it.

However, this need users to write a lot of boilerplate codes, therefor, I think we can add a builtin struct with implement this trait, and which hold a flags and other informations, this can be used for users, also can be used to implement #143.

[shahab96]

Interesting idea. I'm unsure how we'd implement this as I'm still getting used to this codebase but I'll do some investigation today and see what I can come up with

[shahab96]

So, looking at the code again, I don't see how we would be able to do this without having consumers implement ToMemcacheValue explicitly. Additionally, it doesn't seem like a very intuitive interface when implementing the trait for some struct, as now you are able to call set and provide the expiration, but not flags as arguments. This becomes a divergence in the interface where some memcached arguments are part of the set function and others are to be implemented by the consumer. At the very least we can provide a more expansive set of functions and put them behind a feature flag. The interface as a whole doesn't seem to comply with the memcached api and imo it seems like we're working around an interface that was incompletely designed in order to avoid breaking changes.