No theory tells you why something is ultimately true, but a theory can tell you "why" something is true in terms of some more basic elements of the theory; for example, we can use Maxwell's equations to derive the fact that electromagnetic waves can only travel at a single velocity, c. Likewise, in quantum electrodynamics the way that charges attract and repel each other as a function of distance is not treated as a fundamental assumption, but instead can be derived from the way charged particles exchange virtual photons. Something similar might be true of a quantum theory of gravity.
Perhaps you should define what you mean by "incommensurable", since it sounds like you're saying that any
theory is "incommensurable" with any other theory that makes different predictions than it in any circumstances, even if the new theory reduces to the old theory in certain limits, limits which cover every circumstance in which the old theory has successfully passed experimental tests (as with Newtonian mechanics and GR). Would you say that any new theory of physics--say, a theory of particle physics that goes beyond the Standard Model--is "incommensurable" with existing theories? If so, presumably the fact that a new theory would be "incommensurable" with existing ones is not actually a good reason to think that no new theories will be found to supplant older ones, unless you think physics is already complete.
But Einstein didn't construct GR because he wanted to explain the precession of the perihelion of Mercury, or any other specific observational problem that didn't fit with Newtonian mechanics (as this page
mentions, in Newtonian mechanics it was already predicted the perihelion of Mercury's orbit would precess somewhat because of the gravitational influence of bodies other than the Sun, and although the known planets weren't enough to explain the amount of precession astronomers hypothesized that there could be another unobserved planet near the Sun which would make up the difference, so it wasn't completely obvious that this observation actually contradicted Newtonian mechanics). Einstein was motivated by purely theoretical concerns like the inconsistency between Newtonian mechanics and special relativity. Would the 1910 version of you say that he "doesn't want any relativistic theory of gravity", since there is no empirical evidence that necessitates it, in spite of the clear theoretical conflict?
I'm not asking you to "prove" it, just give me a reason to think it is improbable in any way, any more so than any other broad hypothesis about new physics beyond what we already know. For example, do you consider "gravitons exist" more implausible than "GR will no longer work precisely at the Planck scale", or "Quantum mechanics will no longer work precisely at the Planck scale"? There seem to be strong theoretical arguments that one
of the last two must be true.
"Manifestation of the curvature of spacetime" is just a word-picture which one is free to discard even in GR itself, as I mentioned before; it would be better to say something like "all gravitational phenomena can be derived from the equations of GR", since these equations don't necessarily have to be interpreted in terms of "spacetime curvature". And as for your "buckets of observations", of course none of them is in conflict with the hypothesis of a theory of quantum gravity which diverges noticeably from GR at the Planck scale but which becomes almost indistinguishable from GR far from the Planck scale.
The theoretical reasons for postulating gravitons go well beyond "best fit with perturbation theory" (in fact I think most physicists expect a complete theory of quantum gravity to be non-perturbative), they include things like the theoretical incompatibility between GR and quantum theory at the Planck scale, the fact that string theory is known to reduce to GR in certain limits even though it wasn't constructed to have anything to do with gravity, and a number of other hints in the direction of a theory of quantum gravity like the Bekenstein bound
on the amount of information that can be contained in a black hole which can be derived from a number of different plausible theoretical assumptions.
I should not have used "a priori impossible" since that phrase has a specific philosophical meaning and this is the philosophy forum...I was thinking of "a priori" more in the statistical sense
of a prior probability we should assign to something in the absence of evidence, and I erred in saying "impossible" when I just meant something like "very unlikely". So again, why do you think the idea of a quantum theory of gravity which reduces to GR in certain limits is even particularly unlikely (more likely to be false than true, say)? If you are really concerned with justifying everything in terms of observational evidence, then since we have no evidence about Planck-scale physics either way, it would make more sense for you to say something like "there is insufficient basis at present for judging this hypothesis to be either
more likely false than true or
more likely true than false".
I don't think physicists are always so careful about using the word "virtual" when referring to the photons in Feynman diagrams, so I disagree that the word photon "commonly refers" only to non-virtual photons, but I suppose this isn't a very important point either way.