I was unable to build 5.06, nor to run scons -h; I got an error I
didn't understand and don't have time to hunt down.  The line
referenced in SConstruct was the same as the same line in 5.08, which
runs it fine...

But I looked at the code, and I see a comment that a certain loop in
turnoff2 doesn't terminate in 0 mode.  I have been using 4 and 12.
When I tried mode 5, it worked!  But only for the oldest note.  If I
hit a noteoff for the second note added, it crashes like before, with
csound's CPU spiking first as before.

So I guess you're right, it's just broken.  I can't tell exactly
what's going on in it, but here's what I believe to be the offending
loop, from OOps/goto_ops.c:

    do {                        /* This loop does not terminate in mode=0 */
      nip = ip->nxtact;
      if (((mode & 8) && ip->offtim >= 0.0) ||
          ((mode & 4) && ip->p1 != p1) ||
          (allow_release && ip->relesing))
        continue;
      if (!(mode & 3)) {
        if (allow_release) {
          xturnoff(csound, ip);
        }
        else {
          nip = ip->nxtact;
          xturnoff_now(csound, ip);
        }
      }
      else {
        ip2 = ip;
        if ((mode & 3) == 1)
          break;
      }
      ip = nip;
    }

Please if anyone has some spare time and spots a solution, I really
want to use this opcode.  Or if anyone has another way to arbitrarily
turn off individual held notes?

-Chuckk