Wrong eph for small mu (<100)

[gomezzz]

It seems, for small mu values (< 100) , errors for the eph are extremely high / computations become numerically unstable?

The following snippet:

import pykep as pk
import numpy as np 

# Pick some test epoch, problem persists for any
test_epoch = pk.epoch(0)

# Define central body elements, problem persists for any
elements = (1,0,0,0,0,0)

# Some r,v, for orbiting Sat
r = [1000,0.1,0.1]
v = [0.1,0.1,0.1]

# Define central body mu, problems exists only for small MU
for test_planet_mu in np.logspace(-5, 5, 11):
    # Define planet
    test_planet = pk.planet.keplerian(test_epoch,elements,pk.MU_SUN,test_planet_mu,1,1,"test")

    # Define orbiting satellite
    test_satellite = pk.planet.keplerian(test_epoch,r,v,test_planet.mu_self,0,1,1,"test_sat")
    
    r_test,v_test = test_satellite.eph(test_epoch)

    # Compute the difference between the two
    r_diff = np.linalg.norm(np.array(r_test) - np.array(r))
    v_diff = np.linalg.norm(np.array(v_test) - np.array(v))
    print(f"For mu = {test_planet_mu:<10} the difference in r and v is {r_diff:<10.4e} and {v_diff:<10.4e}")

produces

For mu = 1e-05      the difference in r and v is 1.8149e+03 and 1.0162e-07
For mu = 0.0001     the difference in r and v is 8.0696e+03 and 5.9027e-07
For mu = 0.001      the difference in r and v is 3.9075e+01 and 2.3072e-07
For mu = 0.01       the difference in r and v is 1.1187e+03 and 3.5445e-05
For mu = 0.1        the difference in r and v is 8.5343e+01 and 1.4133e-03
For mu = 1.0        the difference in r and v is 2.3625e+03 and 9.1964e-03
For mu = 10.0       the difference in r and v is 6.3135e+03 and 6.9916e-02
For mu = 100.0      the difference in r and v is 1.5194e-12 and 5.9691e-16
For mu = 1000.0     the difference in r and v is 1.2010e-11 and 6.8129e-14
For mu = 10000.0    the difference in r and v is 2.4689e-11 and 1.6205e-12
For mu = 100000.0   the difference in r and v is 1.0816e-09 and 7.2581e-10

@darioizzo maybe this is expected as numerical instability? But then a warning when creating keplerian planets with small MU would be really nice, I think?

(This is pykep 2.6 from conda on windows fwiw)