Navigating tradeoffs in bidirectional filter design

[monkyyy]

a unidirectional range filter can be implimented as simply as:

```d
auto find(alias F,R)(R r){
	while( ! r.empty && ! F(r.front)){
		r.popFront;
	}
	return r;
}
unittest{
	counter(5).find!(a=>a==3).summery;
}
auto findnext(alias F,R)(R r){
	r.popFront;
	return r.find!F;
}
auto filter(alias F,R)(R r){
	struct Filter{
		R r;
		auto ref front()=>r.front;
		void popFront(){r=r.findnext!F;}
		bool empty()=>r.empty;
	}
	return Filter(r.find!F);
}
unittest{
	counter(10).filter!(a=>a%3).summery;//[1, 2, 4, 5, 7, 8]
}
```

Im aware phoboes takes ....more complexity to implement filter 
and there should be *some* extra complexity to make a 
bidirectional filter, but when I ussally look into phoboes 1000 
line functions I usually make different tradeoffs. What are the 
actual considerations here?