stochastic failures
https://github.com/TAPPorg/reference-implementation/actions/runs/21034404114/job/60478289558#step:10:64
https://github.com/TAPPorg/reference-implementation/actions/runs/21034404114/job/60481589441#step:10:64
and subsequent successes (upon simple rerun)
https://github.com/TAPPorg/reference-implementation/actions/runs/21034404114/job/60483956098#step:10:81
seems to be due to BLIS' make/autotools(?)-based build happening inside TBLIS CMake harness. Maybe timestamp corruption? @devinamatthews any ideas? Does BLIS have a CMake harness?
P.S. interestingly, the successful build builds blis after tblis sources and produces different messages
https://github.com/TAPPorg/reference-implementation/actions/runs/21034404114/job/60483956098#step:10:91
this is likely unexpressed dependencies issue in BLIS or (less likely) TBLIS, perhaps can be worked around by disabling parallel build?