fix joblib.load from stream

[PetroFrolov]

I founded an issue when loading data from stream.
Code example

>>> import numpy as np
>>> a = np.array([1, 2, 3])
>>> b = np.array([3, 4, 5])
>>>
>>> import io
>>> import joblib
>>> buf = io.BytesIO()
>>>
>>> joblib.dump(a, buf)
>>> joblib.dump(b, buf)
>>>
>>> buf.seek(0)
0
>>> joblib.load(buf)
array([1, 2, 3])
>>> joblib.load(buf)
array([1, 2, 3])
>>>

VS

>>> import numpy as np
>>> a = np.array([1, 2, 3])
>>> b = np.array([3, 4, 5])
>>> 
>>> import joblib
>>> with open('1.bin', 'wb') as fd:
...     joblib.dump(a, fd)
...     joblib.dump(b, fd)
... 
>>> with open('1.bin', 'rb') as fd:
...     joblib.load(fd)
...     joblib.load(fd)
... 
array([1, 2, 3])
array([3, 4, 5])

My patch will fix it

[codecov]

Codecov Report

Merging #1167 (cbcb397) into master (457d2c8) will increase coverage by 0.08%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #1167      +/-   ##
==========================================
+ Coverage   94.48%   94.57%   +0.08%     
==========================================
  Files          47       47              
  Lines        6997     7000       +3     
==========================================
+ Hits         6611     6620       +9     
+ Misses        386      380       -6     

Impacted Files	Coverage Δ
joblib/numpy_pickle_utils.py	92.77% <100.00%> (+0.08%)	⬆️
joblib/test/test_numpy_pickle.py	93.38% <100.00%> (+0.02%)	⬆️
joblib/_parallel_backends.py	94.92% <0.00%> (+1.17%)	⬆️
joblib/test/testutils.py	100.00% <0.00%> (+50.00%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 457d2c8...cbcb397. Read the comment docs.

[jjerphan]

Hi @PetroFrolov,

Thank you for your pull request.

Can you add your first snippet as a non-regression test?