1. You're talking about the concept of "fatigue limit", the point of stress loading on the s-n curve below which no fatigue occurs. But this is "high-cycle" concern, out at 10 million, 100 million cycles and beyond. To me, that makes the issue basically irrelevant for bicycles, since everything I've read says they fail due to "low-cycle" fatigue.
2. "Low cycle" refers to high loads that cause fatigue failures fairly quickly, like in 10,000 to 100,000 cycles. For bikes, this would be major bumps and potholes, as well as heavy climbing out of the saddle. I'm not an expert on fatigue theory, and can't explain the different mechanisms that distinguish "low-cycle" from "high-cycle" fatigue, but what I've read says that "low-cycle" loading is what kills bike frames and forks, not any very small high-cycle loads, eg like road buzz.
3. Your concern about crack propagation rates is a good one. In aerospace applications, having adequate time to detect impending failures in CF is an important design and certification consideration. Obviously, no one wants an aircraft which has parts that can fail suddenly, either CF or aluminum, steel or ti.
Just an anecdote, but the only guy I know that has fatigued two Trek CF frames hasn't crashed either time. The cracks gave him plenty of warning and time to get home and take the frame to the dealer. Agree if a frame or fork doesn't give an indication and just snaps, that would be a poor and dangerous design. But, with the millions of CF forks now being ridden, abused and crashed everyday, believe they must be pretty well designed and durable. If not, riders would be thrown to the ground regularly by CF forks, and I just don't think that's happening in the real world. After all, the CF itself has high tensile strength, so even if the resin is cracked, the fibers should serve to hold everything together for a good while.
I continue to bring up CF forks because that's the most critical application, IMO. If you're willing to trust your life to a CF fork, the frame material shouldn't matter much.