Make Tangram::Map implementation private

[matteblair]

Instead of holding all of our subsystem objects directly in the Map class, Map now holds a unique_ptr to an object of the private class Map::Impl and all of the subsystem objects are members of Map::Impl.

This significantly reduces the declarations needed in tangram.h and in the future will allow us to change the implementation of Map without changing this header.

[karimnaaji]

Could we clear tiles for all map instance when triggering a debug flag that requests tile reloading?

[matteblair]

Maybe! We'd have to register all map instances to a global variable somewhere, which seemed undesirable.

I had a weird idea on fixing this: getDebugFlag() should remain a global because it means we can use these test conditions from anywhere in the code, and that's really useful. But setDebugFlag() could be a non-static function in Map so that we have access to the tile sets whenever we set a flag that needs to refresh them. It's weirdly asymmetrical, but kinda fits our needs.

[tallytalwar]

Reviewing this now.

[tallytalwar]

Just a note here that Map::Impl class has its move constructor suppressed, unless explicitly defined/defaulted. This might have an affect if the user of the Map class tries to move an instance.
Reference: http://en.cppreference.com/w/cpp/language/pimpl

[tallytalwar]

On similar note, should we restrict copying of a Map object by explicitly deleting copy constructor? RFC!

[matteblair]

Map is implicitly non-copyable - the compiler can't generate a copy constructor because of the unique_ptr

Supporting move semantics for Map would be trivial, I don't know if there's a serious need for it though.

[tallytalwar]

Yeah I also do not see a serious need or use case for a move for Map. Wanted to bring this out, just in case, someone has a use case in mind.

[tallytalwar]

This looks good to me. If no one has further comments will merge (squash, to save one merge commit) this.

[matteblair]

Thanks! Yeah, squash makes sense to me.