BUG: Avoid inplace modification of input array in newton

[Kai-Striega]

optimize.newton previously changes the users input array when
performing calculations. This PR changes this to perform an explicit
copy of the array before any inplace modifications are performed.

closes #9964.

[rgommers]

Hmm, that looks like a recent regression, hard to believe that newton always did that.

[rgommers]

there's a minor style issue that caused CI to fail

[Kai-Striega]

@rgommers fixed the codestyle issue. Looking at the git history, it seems to have been around for a while.

(sp-dev) kai@t470:~/Develop/scipy$ git blame -L 362,366 HEAD~2 scipy/optimize/zeros.py
e1d50a7b7e (Mark Mikofski 2018-06-24 18:21:29 -0700 362)     try:
e1d50a7b7e (Mark Mikofski 2018-06-24 18:21:29 -0700 363)         p = np.asarray(x0, dtype=float)
998a458dd8 (Ralf Gommers  2018-06-24 23:34:28 -0700 364)     except TypeError:
998a458dd8 (Ralf Gommers  2018-06-24 23:34:28 -0700 365)         # can't convert complex to float
e1d50a7b7e (Mark Mikofski 2018-06-24 18:21:29 -0700 366)         p = np.asarray(x0)

[rgommers]

Looking at the git history, it seems to have been around for a while.

Well, those commits went into 1.2.0, so it's a regression in the latest release because of gh-8357.

[rgommers]

Merged, thanks @Kai-Striega

[rgommers]

@mikofski FYI in case you didn't see it

[tylerjereddy]

This is in the 1.3.x maintenance branch already too--presumably backport label is for 1.2.x LTS. I think that gels with the discussion above.