`bench_blas.py::test_gesdd` in Codspeed CI job

[rgommers]

This failure just showed up in CI, from this log:

 benchmarks/bench_blas.py ............................................... [ 75%]
  .........F.....                                                          [100%]
  ================================ 62 benchmarked ================================
  
  =================================== FAILURES ===================================
  ______________________________ test_gesdd[mn1-s] _______________________________
  
  benchmark = <pytest_codspeed.plugin.BenchmarkFixture object at 0x152f7980>
  mn = (1000, 222), variant = 's'
  
      @pytest.mark.parametrize('variant', ['s', 'd'])
      @pytest.mark.parametrize('mn', gesdd_sizes)
      def test_gesdd(benchmark, mn, variant):
          m, n = mn
          rndm = np.random.RandomState(1234)
          dtyp = dtype_map[variant]
      
          a = np.array(rndm.uniform(size=(m, n)), dtype=dtyp, order='F')
      
          gesdd_lwork = ow.get_func('gesdd_lwork', variant)
      
          lwork, info = gesdd_lwork(m, n)
          lwork = int(lwork)
          assert info == 0
      
          gesdd = ow.get_func('gesdd', variant)
          u, s, vt, info = benchmark(run_gesdd, a, lwork, gesdd)
      
  >       assert info == 0
  E       assert 12 == 0
  
  benchmarks/bench_blas.py:237: AssertionError
  =========================== short test summary info ============================
  FAILED benchmarks/bench_blas.py::test_gesdd[mn1-s] - assert 12 == 0
  ================== 1 failed, 61 passed in 2080.74s (0:34:40) ===================
   ** On entry to SLASCL parameter number  4 had an illegal value
   ** On entry to SLASCL parameter number  4 had an illegal value
   ** On entry to SLASCL parameter number  4 had an illegal value
   ** On entry to SLASCL parameter number  4 had an illegal value

This benchmark was introduced in gh-4678 one and a half months ago. I'm not sure it failed before like this. @ev-br you may want to look into this?