[krbdev.mit.edu #3084] need more thorough testing for conditional pthread support

[Ken Raeburn via RT]

When the pthread support may or may not be linked in, either because
we're using weak symbols or because the system C library has stubs for
all the functions we use, we need to do a better job of testing for
*real* pthread support (i.e., not non-functional stubs).

Recommended approach: Test all the symbols we use (and maybe some others
we know of) for null addresses.  Try using pthread_once and see if it
works.  Only if all tests pass should we assume thread support is available.