I'm really confused about inertial vs non-inertial frames and fictitious forces. (...)
First of all, let me clarify something that Ynot had trouble with. GR can tell you what *would* happen in various counterfactual worlds. GR can tell you what an utterly-empty, finite-size Universe would be like---even though we don't happen to live in such a Universe.
By analogy, SR can tell you what *would* happen if you had a relativistic racecar on a relativistic spaceship full of identical twins, even though we don't have such a thing; and questions about "what does SR mean" can be answered in terms of what it says about the spaceship.
1. GR states that there are no inertial frames.
No. In space with masses in it, there's no
global inertial frame.
2. GR recognizes gravity as a consequence of the curvature of spacetime rather than a "force" in the newtonian sense. Gravity is a fictitious force?
Yes. The funny laws-of-motion seen under gravity ("this cannonball's coordinates describe a parabola") come from the exact same GR terms as the funny laws-of-motion ("things get pulled right off the carousel") seen by accelerating observers. There is no identifiable theoretical or experimental distinction that lets you label one as a "force" and one as "fictitious".
Mass/energy causes spacetime curvature, that's a real thing, but
feeling a force due to curved space is, in some sense, a fictitious force.
3. An inertial frame is an approximation of reality. To presume an inertial frame is to disregard (not recognize) the effects of massive objects on the geometry of spacetime.
4. A non-inertial frame is an accurate depiction of reality. Choosing a non-inertial frame is an attempt to recognize the effects of massive objects on the geometry of spacetime.
No. If you're looking at a small, finite region of space, your choice of frame
can fully take into account a nearby gravity/mass/curvature. That's true whether you pick an inertial or non-inertial frame.
The only inaccuracy that comes in is if you try to extend your frame over a large region. Then, the thing you're ignoring is
tidal forces, i.e. gravity gradients, i.e. curvature.
Examples: near Earth, an astronaut in a free-falling space capsule will think of his walls as an inertial frame. A person in an elevator on the Earth's surface will think of his walls as a non-inertial frame. Both of these are perfectly "taking into account" the presence of gravity/curvature. Both are limited, in the same way, by the finite size of the space considered---if you have a hundred-mile-wide box, you might measure that gravity points in a different direction at the left edge of the box vs. the right edge, but that's a tidal force---there's no frame-transformation you can do, inertial or otherwise, to prevent that difference from being there.
A better place to start would be: free-falling observers are, by definition, in inertial frames. Gravity/curvature tells us that sometimes a free-falling observer doesn't move in a straight line (or, equivalently, that an observer moving in a straight line---like one standing on the Earth's surface---is not inertial). If you wish to represent all non-straight-line motion
as though due to a force---as though straight-line motion was the default, and only F=ma could violate it---then you can invent an explanation along these lines.