pickling a dynamic module remove its __builtins__ as a side effect

[pierreglaser]

Reproducer below:

(test) ~/repos/cloudpickle (master)❯_ ipython
Python 3.9.2 | packaged by conda-forge | (default, Feb 21 2021, 05:00:30)
Type 'copyright', 'credits' or 'license' for more information
IPython 7.22.0 -- An enhanced Interactive Python. Type '?' for help.

In [1]: from types import ModuleType

In [2]: m = ModuleType('mod')

In [3]: exec("""
   ...: def f():
   ...:     import math
   ...:     return math.sqrt(4)
   ...: """, vars(m))

In [4]: m.f()
Out[4]: 2.0

In [5]: import cloudpickle

In [6]: roundtripped_m = cloudpickle.loads(cloudpickle.dumps(m))

In [7]: roundtripped_m.f()
Out[7]: 2.0

In [8]: m.f()
---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
<ipython-input-8-ada232509f74> in <module>
----> 1 m.f()

<string> in f()

ImportError: __import__ not found

The issue comes from the _module_reduce function, that removes the item corresponding to the __builtins__ key from the module being pickled (see #325). We should instead copy the module state before removing the item corresponding to the __builtins__ key.