Bug now clearly identified!
The verdict is in. It IS a bug in SampleTank that has been the source of my problems. Let me explain this in as much detail as I can in order to eliminate any needless debate about the credibility of my assertion.
I have a set of 13 samples (which you can download
here if you feel like torturing yourself). All thirteen samples have the root note of D3 and are labeled accordingly. Strict adherence to IK's instructions on how to import samples has been followed. When building the instrument, you must select "VELOCITY FROM NAME". The goal is to have all 13 samples loaded on the D3 key with their range spanning the entire keyboard, the loudest sample triggering at velocity 127, the next loudest at 126, etc., and finally apply stretch mode to maintain realism and playability across the keyboard.
So what is the bug? Answer: You can't load more than 8 samples at various velocities to the same root note. Go ahead and try any combination of 8 of the 13 samples (you must include the sample labeled 127 because SampleTank can't build the instrument with out a sample on 127). The resulting instrument will be stable. But try to load any combination of 9 or more samples in the fashion described above, and the resulting instrument will crash SampleTank stand-alone or the host application (if you are running SampleTank as a plugin) as soon as you start playing the loaded instrument. Mystery solved. You would think you could import 128 velocity layers if you wanted to (0-127), but you can't. You can import only 8.
I want to thank all respondents who offered constructive advice on how to narrow this issue down-especially Eli. In a way I'm relieved because at least I can build instruments with 8 velocity layers. I'm disappointed that IK never made the effort to investigate this issue when I first drew it to their attention. They blamed the samples when it was never the samples at fault. This experiment proves why it never made a difference whether the samples were converted from AIFF to WAV or were exported from the original loop from 3 different audio editors. My hope now is that IK will address this bug and release an update.
Something funny: with respect to the final instrument I was able to create, the final joke is on me: My whole objective in assembling this first instrument was so I could apply Stretch technology to a set of guitar samples, but the overdrive harmonics are so strong in the highest velocities that Stretch cannot make heads or tails of the correct pitch. As soon as I apply any amount of Stretch to the instrument, the pitch on the strongest velocities goes haywire. C'est la vie!
Final note: Don't apply Stretch to samples with a lot of harmonic content and don't try to build instruments with more than 8 velocity layers until that bug is fixed.
PS: I want to apologize in advance if my tone in this post sounds at all smug, pious or otherwise offensive. It's just that I have waited well over a year to get to the bottom of this, and now that I have I'm a little giddy. I hope this research helps someone other than me. And I hope IK admits to and fixes the velocity import limitation. Wow, what an evening! Lastly, my thanks to dfloik, the IK rep, for your understanding and helpful tact. I hope you can forward these results to the developers so this undocumented limitation can be addressed. Btw, I'd still love to be able to save over my own primary instrument names with modified versions, but you can consider that a feature request rather than a bug complaint.