Annarchy

* ANNarchy uses now the CMake build system.
* Added global `monitors()` / `Network.get_monitors()` to iterate across instantiated monitors.
* Added saving of recordings as h5py files (see PR 33)
* Updated heuristics, e.g., selection between SELL/ELL-R, post-to-pre always for continuous signal transmission
* Improved report generated by ANNarchy, e.g., Input section
* Improved unittests (PR 29, thanks to Alex Schwarz)
* Limited the availability of diagonal formats (see commit 461e998)
* Several bugs in the code generation related to non-default matrix formats (e.g., 5b768f2, d4e06d5, a460a83, 3453be6)

* The random distributions now use `np.random.default_rng(seed)`. The default seed is now `None` instead of `-1` for compatibility.
* License of C++ files is GPLv2 or later, as the rest of the code.
* Bugfix 22 by penguinpee: now works with sympy >= 1.13.
* Bugfix 26 on reordering of dendrites in `connect_with_function()`.

* Important: ANNarchy will require Python 3.10.
* Added the diagonal format for rate-coded models using single- or multiple CPU cores.
* Added an `InputArray` population to represent `r` and keep it recordable.
* Refactored the docstrings and added type annotations.
* Fixed the ANN-to-SNN conversion extension for newer tensorflow versions.
* Fixed the use of built-in functions like `pos()` in `report()`.
* Fixed a bug when loading save files resulting from models using structural plasticity (commit 0c13da3, originated in 281e272)
* Progress bar shown when using simulate(progress_bar=True) now uses tqdm.
* Global get_projections() and Network.get_projections() work now similarly (issue 19)
* Fixed a bug for spiking models using structural plasticity (issue 12)
* Fixed a bug when PopulationViews result in IndividualNeurons (commit d2648c8d)

* Removed deprecated "weightsharing" extension (replaced by "convolution" since 4.6.8 release).
* Added a new extension sub-module for ANN-to-SNN conversion. See examples/ann_to_snn for more details.
* Added convolution/pooling on CUDA devices.
* simulate_until() accepts now also a spiking condition (298a598)
* Added two new evaluation methods to Monitor: inter_spike_interval, coefficient_of_variation
* Added binomeal distribution to the set of supported random distributions in ANNarchy (7a667b)
* Extended Monitor.histogram to compute a histogram for neurons individually.
* create_synapse / prune_synapse part of the structural plasticity accepts now a list of ranks.
* TimedArray: the data buffer can now be changed using *update()* after *calling compile()*, see commits 46fe14e, ad8bbbe, 630765f, 3498101
* Refactored CUDA code generation to be compatible with recent NVidia SDK >= 11.x
* Fixed potential memory leaks in Pooling/Convolution/BOLD extension (c302041, 74e75a3)
* Fixed potentially too high memory allocation for SELL, CSRC, and BSR (a3c158)
* Refactored some parts of the CUDA code generation (e.g., f64375b, 7fb07b4)
* Updated unittests (f3d3109, 9283dc3)
* Issue 10/11: bugfix related to psp-statements on CUDA devices
* Removed a case of unnecessary re-compiles of the simulation core (58dcd85)
* PopulationViews support negative indices (bf71f55)
* Refactored command line arguments (e03c28b), e.g., added --auto-tuning (still experimental!) and --report

* Support for Cython 3.0.
* Bugfix: venv does copy python-config.

* Bugfix: Projection.dendrite iterator was not working inside a network.