SuperLU 4.3 called some functions returning floats without declaring
them. On i386@linux call convention, this fails to clear floating point
registers after call. As a result, NaN can appear in the next floating
point operation made.
Squashes a couple of the current test failures on i386.
BUG: SuperLU: fix missing declaration of dlamch
Coverage remained the same when pulling 4796e0d on pv:superlu-bug into 77a7e48 on scipy:master.
Fixes the failure, merging. Thanks @pv