Pdfflow

The main update of this version is the usage of `lhpdf-management` instead of LHAPDF in preparation for 1.3

What's Changed
* Remove LHAPDF non-python dependencies by scarlehoff in https://github.com/N3PDF/pdfflow/pull/57
* Packaging by marcorossi5 in https://github.com/N3PDF/pdfflow/pull/59
* Vectorized example C-API by scarrazza in https://github.com/N3PDF/pdfflow/pull/61
* Add a workflow to test some number of PDFs by scarlehoff in https://github.com/N3PDF/pdfflow/pull/62

**Full Changelog**: https://github.com/N3PDF/pdfflow/compare/v1.2.1...v1.2.2

Minor release of VegasFlow

License is changed from GPLv3 to Apache 2 for consistency with other projects.
Now the environment variable `PDFFLOW_DATA_PATH` can be used to point to the pdf sets so `lhapdf` is no longer needed (55)

Alongside with TF 2.5, python 3.9 is now supported.

A list of members (or `None` at all) can now be given to PDFFlow (50) making the interface a bit more intuitive.

Minor update of PDFFlow. The main feature with respect to the 1.0 release is that we have dropped the dependence on `tensorflow-probability` and added a explicit example of an interface of PDFFlow with Fortran and C++

First production release of ``PDFFlow`` to accompany the upload of the paper to the [arXiv](https://arxiv.org/)

This first release contains the basic ingredients for PDF interpolation in hardware accelerators as described in the [documentation](https://pdfflow.readthedocs.io/en/latest/?badge=latest) and the upconming paper.

The main features are:

- Parallel PDF interpolation in GPU and CPUs
- Parallel alpha_s interpolation in GPU and CPUs
- TensorFlow-only and Python interfaces available

Release candidate to test that everything works as it should.