- Add an optional CMake option `ARB_CXX_FLAGS_TARGET` for setting target-specific flags to use when compiling for the target architecture (not applied to compilation of modcc).
- If `ARB_ARCH=="none"` CMake will not add architecture-specific `mtune/march/mcpu` flags
- Remove `ARB_CXXOPT_ARCH` from the installed `arbor-config.cmake`, and replace with more general `ARB_CXX_FLAGS_TARGET`.
- Update spack `package.py` to use this feature to pass custom flags.

Fixes #1519
Fixes #1522 

Replaces PR #1518 

Bump version for release

* Turn each task queue in the task system into a finite set of queues of increasing priority. The number of queues is a compile-time constant, currently two.
* When a task of priority higher than that of the highest priority queue is scheduled, execute it synchronously.
* When scheduling tasks in a task group, use a priority one higher than that of the calling task; when waiting on the task group, work only on tasks with this priority or higher. This is sufficient to avoid the deep recursion issue seen in issue #1570.

Fixes #1570.
Co-authored-by: Sam Yates <yates@cscs.ch>

* Skip relevant corner cases in the matrix solver.
* Minor clean-ups in `make_span`.
* Add regression test for bug, additional unit tests for `span` and `make_span`.

Closes #1540.

New structs and types:
* `cell_tag_type` (std::string): for labelling placeable items on a cell. The label refers to a number of items placed on a locset, equal to the number of locations in a locset. The number of locations in not always known to the user, so the previous way of using indices for items was no longer sufficient. 
* `lid_selection_policy`: for allowing a user to select a single item from a group of items sharing a label. Currently only `round_robin` and `assert_univalent` are supported. 
* `cell_local_label_type` and `cell_global_label_type`: for identifying the target and source of a connection or gap_junction connection. 
* `cell_label_ranges`, and `cell_labels_and_gids`: for propagating information about the labelled items on the cell from the cell groups back to the simulation and communicator. 
* `label_resolution_map` and `resolver`: for selecting an item (and retaining state) from a labelled group of items on a cell according to a user-selected policy.

Changes to the model-initialization: 
* The `communicator` now needs `label_resolution_maps` constructed from the cell group data in order to build the `connections` vectors. 
* The `simulation_state` object handles the transfer of the information when it is constructed. 
* Spike exchange at runtime remains unchanged, because `communicator::connections` remains unchanged. 

Changes to cells, cell_groups and recipe:
* `decor::place` expects a third label parameter, no longer returns an `lid_range`.
* `lif`, `source`, and `benchmark` cells need source/target labels in their constructors. 
* A `cell_group` needs to save data about the gid/labels/lid_ranges of each cell, to propagate back to the `communicator` constructor. 
* Connections/gap junction connections are formed between {label, policy} pairs on cells instead of indices. 
* `num_sources`, `num_targets`, `num_gap_junction_sites` deleted from `recipe`.

Additional changes:
* Add MPI wrapper for exchanging vectors of strings. 
* Corresponding updates to unit tests, Python wrapper, C++ and Python examples, documentation.

Fixes #1394

* Sort collected spikes in Python API lfirst by time, then gid, then lid.
* Remove redundant old spike handling code from Python lib.

* Delete symlinks from repository for Windows devs
* Dropped extension of executable scripts to follow shell convention
* set executable bits in scripts/

Co-authored-by: Robin De Schepper <robin.deschepper93gmail.com>
Co-authored-by: Brent Huisman <brent@huisman.pl>

* Raise KeyError in catalogue.
* Raise KeyError in label_dict.

The C++ label parse code was returning the error type in the value part of a `parse_hopefully<std::any>`,
and not in the error slot.

The `label_dict` Python wrapper caches label strings and their descriptions in order to
provide a convenient Python iterator interface and `[]` access. These were not being
properly updated or set when the label_dict was constructed from a C++ label_dict.

* Update URLs to arbor-sim.org
* update badge links

Add a CI test for ARM-SVE

- compiles and runs the unit tests with vectorization enabled for the `armv8.2-a+sv` architecture
- uses a dedicated gitlab runner on ault.cscs.ch

Fixes #1498

Fixes issue #1529.

* Adapt CSS for the HTML5 transition in docutils 0.17
* Fix some reST markup errors
* Add meta-tag to enable indexation control at Google (current indexation appears at least a year old).

Fix mistakes reported by @Helveg in the Python tutorial docs. 

* Add a GH workflow to build and run Arbor tests and examples with Clang sanitizers.
* Fix some uninitialized struct members; missing `override` qualifiers.

* Use a struct with meaningful field names instead of a `std::pair` to represent elements in `pw_elements`.
* Describe explicitly the semantics of the zip operation.
* Update implementation and add unit tests to suit.
* Add comment explaining `pw_elements::equal_range`.
* Remove parameter names from unused parameters in `impl::piecewise_pairify`.

* Correct bounds check order in communicator.
* Use `.at()` in test recipe to catch failures in bound checking.

* Exit with correct exit code from Python test runners.
* Fix unit tests that were using older APIs.

* Remove std::filesystem to workaround OSX features; use `std::string` in interfaces expecting a path.

Set option `-mcpu=aarch64` for target `aarch64-apple-darwin`

In response to:

https://gizmodo.com/u-s-federal-investigators-are-reportedly-looking-into-1846707144

Rationale:
* CodeCov didn't detect the breach for two months
* They took a further 2 weeks to alert users after that
* Their efforts to address the issue after that have been painful to watch
  * https://github.com/codecov/codecov-action/issues/281#issuecomment-823407167