Test failure for 1.1.1p - 10-test_bn

[rsbecker at nexbridge.com]

Our test process for 10-test_bn went into a hard loop when building using
IEEE float. This has not happened in prior tests or when using platform
floating point (default). The situation does not occur in 3.0.4 - only
1.1.1p. At 1.1.1o there was no issue. I am concerned that there are
assumptions made relating to BN processing with this change but do not know
enough about the Montgomery structures to be able to isolate the cause of
this.

This happens on the big-endian NonStop x86 platform.