Use oldest supported numpy version during build

[tykurtz]

Fixes an issue related to #151 .

To reproduce

mkvirtualenv temp_venv --python=python3.7
pip install toppra==0.4.0 numpy==1.18.5  # or pip install -e . on upstream/master
python -c "import toppra"

and you should get

Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/home/tyler/.virtual_envs/bar/lib/python3.7/site-packages/toppra/__init__.py", line 16, in <module>
    from . import constraint
  File "/home/tyler/.virtual_envs/bar/lib/python3.7/site-packages/toppra/constraint/__init__.py", line 83, in <module>
    from .linear_joint_velocity import (
  File "/home/tyler/.virtual_envs/bar/lib/python3.7/site-packages/toppra/constraint/linear_joint_velocity.py", line 3, in <module>
    from toppra._CythonUtils import (_create_velocity_constraint,
  File "toppra/_CythonUtils.pyx", line 1, in init toppra._CythonUtils
ValueError: numpy.ndarray size changed, may indicate binary incompatibility. Expected 88 from C header, got 80 from PyObject

Changes in this PRs:

• Fixes a build issue against older versions of numpy

Checklists:

• [x] Update HISTORY.md with a single line describing this PR

Some background on what exactly this change is can be found here, https://pypi.org/project/oldest-supported-numpy/

The reason to use the oldest available Numpy version as a build-time dependency is because of ABI compatibility. Binaries compiled with old Numpy versions are binary compatible with newer Numpy versions, but not vice versa.

and https://github.com/scikit-learn-contrib/hdbscan/issues/457#issuecomment-773671043

Probably, pip has always been doing this and the only reason it's breaking now is because numpy==1.20.0 is a breaking ABI change. This also means that there's nothing to "roll back", because the mere existence of a new numpy is enough to cause it.

The other way around this would be to require numpy>=1.20 but that would rule out setups which require older versions of numpy (notably tensorflow)

[dgoel]

Friendly ping to @hungpham2511 for review.

[hungpham2511]

Hey thanks for the PR and, pinging me. Have been kind of busy recently and sort lost track of this PR. I will have a look shortly :)

[dgoel]

LGTM. Thanks!

Thank you so much! Could you publish a new release to pypi ?

[hungpham2511]

I just kick-start the CI. A new version should be available soon.