[32-bit Linux] Investigate Linear Algebra failure

[mratsim]

https://github.com/mratsim/Arraymancer/pull/510/checks?check_run_id=2980291716#step:21:823

The first failure may be due to rounding issue with float32 and switching to float64 will solve the issue.
1e-11 is way over float32 precision.

Arraymancer/tests/linear_algebra/test_linear_algebra.nim

Lines 130 to 147 in 8514e78

	block: # p15 of http://www.cs.otago.ac.nz/cosc453/student_tutorials/principal_components.pdf
	let a = [[0.616555556'f32, 0.615444444],
	[0.615444444'f32, 0.716555556]].toTensor
	let expected_val = [0.0490833989'f32, 1.28402771].toTensor
	# Note: here the 1st vec is returned positive by Fortran (both are correct eigenvec)
	let expected_vec = [[-0.735178656'f32, -0.677873399],
	[ 0.677873399'f32, -0.735178656]].toTensor
	let (val, vec) = symeig(a, true, 'U')
	check: val.mean_absolute_error(expected_val) < 1e-7
	for col in 0..<2:
	check: mean_absolute_error( vec[_, col], expected_vec[_, col]) < 1e-11 or
	mean_absolute_error(-vec[_, col], expected_vec[_, col]) < 1e-11

The second issue is very strange though

Arraymancer/tests/linear_algebra/test_linear_algebra.nim

Lines 300 to 326 in 8514e78

	test "LU Factorization":
	block: # M > N
	# import numpy as np
	# from scipy.linalg import lu
	# np.set_printoptions(suppress=True) # Don't use scientific notation
	# a = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9], [10, 11, 12]])
	# pl, u = lu(a, permute_l = True)
	# print(pl)
	# print(u)
	let a = [[ 1.0, 2, 3],
	[ 4.0, 5, 6],
	[ 7.0, 8, 9],
	[10.0,11,12]].toTensor()
	let expected_pl = [[0.1, 1 , 0],
	[0.4, 0.66666667, 0],
	[0.7, 0.33333333, 1],
	[1.0, 0 , 0]].toTensor()
	let expected_u = [[10.0, 11.0, 12.0],
	[ 0.0, 0.9, 1.8],
	[ 0.0, 0.0, 0.0]].toTensor()
	let (PL, U) = lu_permuted(a)
	check:
	PL.mean_absolute_error(expected_pl) < 1e-8
	U.mean_absolute_error(expected_u) < 1e-8