TST: seed the starting point in test_rotated_hyper_ellipsoid #993

TestMaximization.test_rotated_hyper_ellipsoid in quantecon/optimize/tests/test_nelder_mead.py draws its starting point from the unseeded global random number generator:

Each CI run therefore starts Nelder–Mead from a different point, and occasionally it stops short of the optimum. It failed on #983, in the tests (ubuntu-latest, 3.13) job of run 38883722176 (attempt 1), with x[4] = -1.9e-4 against atol=1e-6, and passed on re-run. Locally, 13 of 30,000 random starting points fail (about 0.04%), with errors up to 1.6e-3. With about a dozen test jobs per push, that is roughly one red run in 200 pushes.

Seeding the draw, or fixing x0, would make the test deterministic. The seed should be one from which the test passes. This is the only unseeded draw in the file.