Fix trajectory_predict_skypos mutating input array

[wilsonbb]

trajectory_predict_skypos is meant to predict the positions of a KBMOD trajectory along a list or array of observation times, however it mutates the observation times if they are provided as a numpy array rather than a list.

It calls dt = np.asarray(times) which creates a new numpy array if times is a list or is a no-op if it is an array. We then subtract out the starting timestamp so all times are offset from 0, and we use the in-place operation of dt -= dt[0] to preserve memory. However if times is an array, the original object passed to the array has been mutated to now be offset by 0.

As quickly illustrated in a python repl:

>>> import numpy as np
>>> def modify_me(n):
...     new_n = np.asarray(n)
...     new_n -= new_n[0]
... 
>>> a = list(range(10, 20))
>>> modify_me(a)
>>> a
[10, 11, 12, 13, 14, 15, 16, 17, 18, 19]
>>> b = np.asarray(a)
>>> modify_me(b)
>>> b
array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9]) 

This PR fixes this by choosing not to do the in-place subtraction.

To add better testing for this, we a) use tests with non-zero observation times to actually test the subtraction and b) test both lists and numpy arrays as inputs.

Note that this was showing up as an issue with the KnownObjMatcher having its numpy array of observation times mutated by the call to trajectory_predict_skypos. Then when it tries to compare the observation times against times from an ephemeris, a match cannot be made due to the temporal subtraction.

[jeremykubica]

Looks good. I added some optional comments (one stylistic and one extending the test).