TST fix platform sensitive test: [all random seeds] test_float_precision

[ogrisel]

Closes #27846.

[github-actions]

✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: 0287379. Link to the linter CI: here}

[ogrisel]

Hum.

assert_allclose(Xt[np.float32], Xt[np.float64], atol=Xt[np.float64].max() * rtol)

fails for seed 33 on the ubuntu atlas build with:

E           AssertionError: 
E           Not equal to tolerance rtol=0.0001, atol=0.00222324
E           
E           Mismatched elements: 180 / 800 (22.5%)
E           Max absolute difference: 0.04510504
E           Max relative difference: 0.02222245
E            x: array([[ 7.528463,  5.820543,  3.035802,  1.845218,  7.03596 ,  6.36431 ,
E                    6.242327,  1.122712],
E                  [ 2.215386,  5.555049, 10.260398,  8.833364,  2.335838,  5.359241,...
E            y: array([[ 7.52846 ,  5.820543,  3.035804,  1.862278,  7.035962,  6.364307,
E                    6.242327,  1.118434],
E                  [ 2.215385,  5.55505 , 10.260399,  8.846413,  2.335837,  5.359238,...

This seems well above the expected floating point rounding errors...

[ogrisel]

I cannot reproduce locally, so maybe it's BLAS-implementation specific.

[ogrisel]

Let's skip this test on Atlas.

[jeremiedbb]

more readable imo

Suggested change

	not [i for i in threadpool_info() if i["user_api"] == "blas"],
	not any(i for i in threadpool_info() if i["user_api"] == "blas"),

[jeremiedbb]

do we still need the increased rtol then ?