Implement dependency injection into workers, resolves #203

[mvertes]

A new skale context method sc.require is added. It specifies a
set of modules on user side (master) which have to be deployed in
workers for use by callbacks, such as mappers, reducers, etc.

Under the hood, browserify is used on master side to build a
bundle which is serialized and sent to workers (as part of task).
It is then evaluated in worker global context, and dependencies
remain persistent as long as workers live.

This method allows to use in workers any javascript module which
can be browserified, so a large number (almost any pure JS package).

The current commit is for local version, not distributed (code will
be exactly the same). It is experimental for the moment.

An additional statement sc.bundle could be added as well to inject
pre-compiled modules, avoiding the penalty of browserify at each
run.

[CedricArtigue]

That's really good. I wonder if we can prevent users from having to explicitly call sc.require by just bundling required modules unconditionally. We can find all non-core required modules with require.cache.

It may be much more natural this way as user shouldn't have to think wether the required module will be used on master or worker side.

[mvertes]

Yes, automatic require could be an evolution once feature is stabilized. First, let fix distributed mode, then validate with real-life complex dependencies, such as kafka, cassandra, etc... which may require tuning browserify and/or worker internals.