Multivar-horner

__________________

* supporting evaluation at complex query points as requested in issue 37

internal:
* hypothesis tests: evaluation equality of Horner and regular polynomials (up to numerical errors)
* added mypy typing pre-commit hook

__________________

* bump pytest dependency version to ``>=7,<8`` (vulnerability fix)
* less strict dependency pinning to support python 3.11
* added python 3.11 tests (not yet supporting numba)
* disabled numerical stability tests due to missing numpy support for high precision 128-bit float on arm64 architecture

__________________

* bump numpy dependency version to ``1.22`` (vulnerability fix)
* officially supported python versions ``>=3.8,<3.11`` (due to numpy and numba constraints)

__________________

* bugfix: packaging. now completely based on pyproject.toml (poetry)

__________________

* bugfix: optional ``numba`` dependency. numba imports were not optional
* bugfix: create __cache__ dir if not exists
* minor documentation improvements
* bumping dependencies

__________________

ATTENTION: major changes:

* introduced the default behavior of compiling the evaluation instructions in C code (C compiler required)
* the previous ``numpy+numba`` evaluation using "recipes" is the fallback option in case the C file could not be compiled
* as a consequence dropping ``numba`` as a required dependency
* added the "extra" ``numba`` to install on demand with: ``pip install multivar_horner[numba]``
* introduced custom polynomial hashing and comparison operations
* using hash to cache and reuse the instructions for evaluation (for both C and recipe instructions)
* introduced constructions argument ``store_c_instr`` (``HornerMultivarPolynomial``) to force the storage of evaluation code in C for later usage
* introduced constructions argument ``store_numpy_recipe`` (``HornerMultivarPolynomial``) to force the storage of the custom "recipe" data structure required for the evaluation using ``numpy`` and ``numba``
* introduced class ``HornerMultivarPolynomialOpt`` for optimal Horner Factorisations to separate code and simplify tests
* as a consequence dropped construction argument ``find_optimal`` of class ``HornerMultivarPolynomial``
* introduced constructions argument ``verbose`` to show the output of status print statements
* dropping official python3.6 support because ``numba`` did so (supporting Python3.7+)

internal:

* using poetry for dependency management
* using GitHub Actions for CI instead of travis