If you like list comprehensions, don't forget about their "lazy" counterpart: generator expressions. These should at least to some extent solve the question of dynamically constructing queries. Complement this with itertools, and of course some of the builtin functions that work on iterables, and you could say you have your Python "LINQ to Object" (or at least something very close).
You won't get "pseudo-sql" syntax built in in python as with C# and LINQ, but that is a matter of taste I guess. (personally, I like the fact that it remains Python, uses the same keywords etc. instead of trying to be sql: familiar if and for clauses instead of introducing where and from and so on).
The object-oriented DSL approach seems feasible (and easier to implement than string-based I think). You will find this sort of thing in ORMs like SqlAlchemy, but I don't know if anyone has done that already for "normal" data structures (I guess people don't really feel it's necessary?)
For indices and so on: I guess you'll have to search for actual "object databases"...
