Make tests more reproducible by seeding based on test paths

[voltjia]

pytest output:

============================= test session starts ==============================
platform linux -- Python 3.12.3, pytest-8.4.1, pluggy-1.6.0
rootdir: /home/huang/ntops
configfile: pyproject.toml
plugins: jaxtyping-0.3.2, typeguard-4.4.4, cov-6.2.1, anyio-4.9.0
collected 636 items

tests/test_abs.py ........                                               [  1%]
tests/test_add.py ........                                               [  2%]
tests/test_addmm.py .F                                                   [  2%]
tests/test_bitwise_and.py ................                               [  5%]
tests/test_bitwise_not.py ................                               [  7%]
tests/test_bitwise_or.py ................                                [ 10%]
tests/test_bmm.py ..                                                     [ 10%]
tests/test_clamp.py ........                                             [ 11%]
tests/test_cos.py ........                                               [ 13%]
tests/test_div.py ........                                               [ 14%]
tests/test_dropout.py ........                                           [ 15%]
tests/test_eq.py ........                                                [ 16%]
tests/test_exp.py ........                                               [ 18%]
tests/test_ge.py ........                                                [ 19%]
tests/test_gelu.py ........                                              [ 20%]
tests/test_gt.py ........                                                [ 22%]
tests/test_isinf.py ........                                             [ 23%]
tests/test_isnan.py ........                                             [ 24%]
tests/test_layer_norm.py ............................................... [ 31%]
.................................................                        [ 39%]
tests/test_le.py ........                                                [ 40%]
tests/test_lt.py ........                                                [ 42%]
tests/test_mm.py ..                                                      [ 42%]
tests/test_mul.py ........                                               [ 43%]
tests/test_ne.py ........                                                [ 44%]
tests/test_neg.py ........                                               [ 46%]
tests/test_pow.py ........                                               [ 47%]
tests/test_relu.py ........                                              [ 48%]
tests/test_rms_norm.py ................................................. [ 56%]
...............                                                          [ 58%]
tests/test_rotary_position_embedding.py ................................ [ 63%]
........................................................................ [ 75%]
........................                                                 [ 78%]
tests/test_rsqrt.py ........                                             [ 80%]
tests/test_scaled_dot_product_attention.py ............................. [ 84%]
.................................................                        [ 92%]
tests/test_sigmoid.py ........                                           [ 93%]
tests/test_silu.py ........                                              [ 94%]
tests/test_sin.py ........                                               [ 96%]
tests/test_softmax.py ........                                           [ 97%]
tests/test_sub.py ........                                               [ 98%]
tests/test_tanh.py ........                                              [100%]

=================================== FAILURES ===================================
__________________ test_cuda[492-430-881-dtype1-0.025-0.025] ___________________

m = 492, n = 430, k = 881, dtype = torch.float16, atol = 0.025, rtol = 0.025

    @skip_if_cuda_not_available
    @pytest.mark.parametrize(*generate_arguments())
    def test_cuda(m, n, k, dtype, atol, rtol):
        device = "cuda"
    
        input = torch.randn((m, n), dtype=dtype, device=device)
        x = torch.randn((m, k), dtype=dtype, device=device)
        y = torch.randn((k, n), dtype=dtype, device=device)
        beta = gauss()
        alpha = gauss()
    
        ninetoothed_output = ntops.torch.addmm(input, x, y, beta=beta, alpha=alpha)
        reference_output = torch.addmm(input, x, y, beta=beta, alpha=alpha)
    
>       assert torch.allclose(ninetoothed_output, reference_output, atol=atol, rtol=rtol)
E       AssertionError: assert False
E        +  where False = <built-in method allclose of type object at 0x77fd92b7bfa0>(tensor([[ -17.7812,   16.5625,  -11.5156,  ...,   83.8750,   35.4375,\n           35.5625],\n        [  69.8750,   38.21...  7.2578,   91.1875,  -16.5469,  ..., -189.5000,  -89.9375,\n           28.7031]], device='cuda:0', dtype=torch.float16), tensor([[ -17.7812,   16.5625,  -11.5156,  ...,   83.9375,   35.4375,\n           35.5625],\n        [  69.9375,   38.21...  7.2461,   91.1875,  -16.5469,  ..., -189.6250,  -89.9375,\n           28.7188]], device='cuda:0', dtype=torch.float16), atol=0.025, rtol=0.025)
E        +    where <built-in method allclose of type object at 0x77fd92b7bfa0> = torch.allclose

tests/test_addmm.py:24: AssertionError
=========================== short test summary info ============================
FAILED tests/test_addmm.py::test_cuda[492-430-881-dtype1-0.025-0.025] - Asser...
================== 1 failed, 635 passed in 633.69s (0:10:33) ===================