Work around numerical instability of the mio N example

[vanderhe]

This slight change is enough to work around the numerical instabilities encountered for the nitrogen STO basis of the mio example. In fact, it is even closer to the geometric series (0.5000 1.2051 2.9044 7.0000) suggested by skexp.

Difference to the original slateratom binary (+original exponents), as reported by skdiff:

*** Atomic data:
Onsite:         2.739e-08     1
Hubbards:       5.742e-07     1
Hubbard (s):    5.742e-07
Occupations:    0.000e+00     0

*** Integral tables:
Hamiltonian:    1.820e-06 ( 104,  9)
Overlap:        4.158e-06 ( 190,  9)

Difference to the original slateratom binary (+updated exponents), as reported by skdiff:

*** Atomic data:
Onsite:         8.272e-09     1
Hubbards:       3.631e-07     1
Hubbard (s):    3.631e-07
Occupations:    0.000e+00     0

*** Integral tables:
Hamiltonian:    3.706e-07 (  19,  5)
Overlap:        2.660e-11 (  45,  6)

Closes #92 (not really a fix though).

[bhourahine]

Seems OK, but would be better to switch to something algorithmically more robust.

[vanderhe]

Seems OK, but would be better to switch to something algorithmically more robust.

Yeah, this PR does not really qualify as a fix. Strictly spoken is not even the mio nitrogen anymore, but what can you do...