Dear Constantin,
As to the eclipse, it is not light which bends around the Sun but spacetime itself which is curved and affects the physical relation between the distant star and the observer. To the photon itself, its path would be perfectly straight if not for the fact that to the photon there is no path, no spacetime distance between the points it is transmitted. If to the photon itself it doesn't interact as it is transmitted, then we cannot insist that it does. Its transmission doesn't consist of three separate, independent events, its emission, its voyage and its (accidental) absorption somewhere else, but is a single event. If A emits a photon which is absorbed by B, a transmission changing the state of both A and B, then A sees the state of B change at the time it emits the photon, whereas B sees the state of A change as it absorbs the photon. Though to us the moment when A sees B change and when B sees A change don't coincide in time, it nevertheless it is a single event taking no time at all. To A its own change doesn't precede that of B, nor sees B change itself before or after it sees A change. Only if we assume that there is a God-Clock, an Absolute Time we can determine in an absolute sense what precedes what, is cause of what, and only then can we interpret the speed of light as a velocity. Without such a clock it is just a property of spacetime, how many meters correspond to how many seconds. In my essay I have repeatedly emphasized that there is no such clock.
As to a photon curving spacetime: according to GR energy only is a source of gravity if and when it can be assigned a position, like I argued in my essay: the energy of a (massive) particle likewise depends on the definiteness in its position. Since the speed of light can be defined as that 'velocity' at which the position of a particle is completely indefinite so it cannot express its energy in interactions, act upon anything, a photon cannot curve spacetime. Another reason why photons cannot, should not interact, is that for photons to be able to transmit force between two particles, they must be impervious to influences which may affect that force, which is why it has no mass, why it isn't delayed by interactions as goes about its business, why it has the 'speed' of light. We should reserve the term 'velocity' for motions proceeding at v < c, for classical situations where causally applies. Though we can describe the propagation of light as if it has a velocity, we should keep in mind that it is a non-causal QM phenomenon. Spacetime is curved by mass, by the continuous energy exchange by means of which particles express and preserve each other's energy, not by photons. This, by the way, is not to say that we cannot predict where and when we can intercept a photon, that is, make the source produce photons for our detector. We should not think about photons as some kind of bullets buzzing through spacetime, as if the state of the source particles can change before that of the absorbing particles, 'before' according to the God-Clock. Unfortunately causality itself cannot exist without such clock, so if we reject Absolute Time (as we must if we are to practice physics instead of metaphysics), then we should discard causality. Though events certainly are related, we shouldn't waste time on irrelevant questions as to what is cause of what. So to say that ''there are some problems with causality in quantum mechanics'' seems to me a gross understatement. I'm well aware that what I propose is completely different from the present consensus about what photons are, so I can imagine your confusion.
I do not describe the EPR paradox with the above AB sketch. In an EPR experiment (like those of Anton Zeilinger) which entangles two photons, the light source (the particles to emit the photons) at all times is informed about the direction of the polarisation filter in the detector as it continuously exchanges energy with all particles of the setup. So if the source keeps producing pairs of entangled photons, then any photon with the right polarisation to pass the detector filter and be detected is produced together with a photon with the opposite polarisation. As the information about the detector already is present at the source at the time of the emission, there's no need for signals traveling instantaneously from the detector to the other photon which then is thought to assume the appropriate polarisation. It already is produced at the source with that polarisation, so no info is transmitted faster than light and there's no teleportation of Q states. The idea of teleportation originates in the belief that there is a God Clock, in the outdated, 19th century idea that speed of light is a velocity.
Regards, Anton