lalapps failing Weave tests on arm64 architecture

After patching lalpulsar's ComputeFstatTest (to tolerate a mismatch of 5e-8 instead of exact zero in the v1-v1 test), I'm now running into serious issues with lalapps's Weave tests. The build logs are about 140 MB in size, too large to be included here. They can be picked up here: https://galahad.aei.mpg.de/lsc-arm64-stretch/lalapps_6.22.0-1+deb9u0_arm64.--pbuilderlog0 https://galahad.aei.mpg.de/lsc-arm64-buster/lalapps_6.22.0-1+deb10u0_arm64.--pbuilderlog0 Search for "mismatch" or "metric dot" as a starting point.

The prominent "feature" of the issue is that "metric dot"'s components exceed 1 by several orders of magnitude, exceeding the total tolerance limit of 0.001 by huge factors. Also I have seen identical indices result in changed values (?).

The same result (with almost identical numbers) has been found both in Stretch and Buster builds on a Cavium ThunderX machine, and could be reproduced on a Raspberry Pi 3B - both running 64-bit Debian (aarch64 aka arm64).

Access to the big machine can be provided via Vulcan. I will create accounts on request (send x509 subject).

Note: issue #85 (closed) might be affected.