Experimental support for mixins with interpolated selectors (#1196).

This is no doubt hackish patch (quick and dirty). However since it introduces no breaking changes I hope it can serve as a temporary solution for #1196, #1399.

What is supported: see included tests.

What is not supported: matching for selectors concatenated across nesting boundary (the same way and for the same reason this does not work with regular mixins), i.e.:

@c: c;
.a {
    &-b    {}
    &-@{c} {}
}

test {
    .a-b; // error, undefined
    .a-c; // error, undefined
}

Furthermore, this patch does not solve any scope visibilty problems mentioned in #1399.

Undocumented feature:
Though this patch allows a variable to contain whatever complex selector it might be and such selector will be matched correctly, for example:

@v1: 42;
@v2: ~".a >.b.c .d + #e~ #f";
.a > .b.c .d + #e ~ #f {
    &.g@{v1}-@{v2}-banana {
        boo: 1;
    }
}

.test {
    .a.b.c.d#e#f.g42-.a.b.c.d#e#f-banana; // OK
}

It would make sense to leave this undocumented and never recommend to use anything like this because such possibility may be not so easy to support in future. So basically, a recommendation is: a variable used in an interpolated selector should not contain anything beyond a single selector element and/or a naked string (e.g. foo, ~".foo" or ~"#foo").