Pyfstat

- Fixed some detection statistic / output values corner cases.
- E.g. in case the user wants `twoFX` (via `singleFstats` option), but not `BSGL`.
- This includes fixes for regressions in 2.2.0.
- Added `t0_MP` and `tau_MP` max-posterior estimates in transient outputs,
following lalpulsar conventions.
- Correspondingly, renamed the max-likelihood values to `t0_ML` and `tau_ML`.
- Added support for `BtSG` detection statistic in `SearchOverGridFile`.
- Improved logging of grid search classes.
- Added `utils.translate_keys_to_lal()` function.

- Changed the core F-stat functions to alternatively take parameter-space points as dictionaries.
The old individual arguments are still supported, but deprecated and will be removed in a future version.
- Added new function `utils.plot_spectrogram`.
- Added support for spindown orders >3 in Writer and search classes (up to `lalpulsar.PULSAR_MAX_SPINS`).
- Added SearchOverGridFile class that can read a grid file following the format of `lalpulsar_ComputeFStatistic_v2`.
- Some internal refactoring of code between mcmc, grid and core classes.
- Updated various dependencies, mostly for developers.
- Pinned chainconsumer dependency to <1.0.
- Added codespell to pre-commit hooks.
- Fixed some deprecations and code-formatting issues.
- Extended tests.
- Changed tutorial notebooks plotting from pcolorfast to pcolormesh.

- This is a maintenance release for updating python and numpy compatibility and fixing minor bugs.
- Dropped python 3.8 support.
- Enabled python 3.12 support.
- Enabled support for numpy >=1.24.0 as long as it's still <2.0.
- Fixed binder notebooks.
- Fixes to some MCMC plotting methods in case of returning fig and axes (rather than saving to files).
- Fix for transient parameters in get_predict_fstat_parameters_from_dict().
- for developers: upgraded to black 24 style,
refactored and updated the github actions

- Pinned to python<3.12 until more dependencies are updated
and a few things fixed on our end.

- Fixed compatibility with matplotlib=3.8 and bumped minimum requirement to >=3.3.
- Fixed pulling in ephemerides via lalpulsar optional dependency.
- Fixed line simulation in "generating noise" tutorial.
- Fixed some type checks to conform to flake8>=1.7 standards.
- PyPI uploads now done using "trusted publishing" (OIDC).

- This is a major release of PyFstat in the sense that
users will likely have to make some changes to the way they call it,
but mostly just regarding class instance labels / file naming.
- Mainly, we follow the recent `lalpulsar` upgrade to v3 of the SFT file format specification
(see https://dcc.ligo.org/T040164-v2/public
and note the v2 in the URL is not a typo,
this file version describes both the v2 and v3 specifications).
- We now require `lalsuite>=7.13`.
- The file format update itself is fully backwards compatible:
the only change is including window information in the header,
which however reuses existing padding bytes, and hence does not affect compatibility.
- The file naming convention however has become more restrictive:
SFT files, and hence `label` arguments for `Writer` and derived classes,
now may only contain ASCII alphanumerical characters,
which specifically makes the old PyFstat habit of using underscores illegal.
We suggest CamelCase instead.
- Instead of `SFTWindowBeta`, one must now use `SFTWindowParam`.
- `noiseSFTs` no longer requires `SFTWindow[Beta/Param]`,
only if a window specification cannot be read from the headers of input SFTs.
- Fixed an error that prevented one parameter to be printed in search output file header comments.
- Fixed compatibility with `corner=2.2.2`.
- Removed `utils.get_lal_exec()`, now always expect executables to be named `lalpulsar_`.
(Old LALSuite versions with `lalapps_` CW executables are quite outdated by now.)
- Removed deprecated prior formats in `InjectionParametersGenerator` class.
- Transient F-stat GPU implementation:
- Introduced F-stat condition number check,
with threshold equivalent to defaults of
`XLALComputeAntennaPatternSqrtDeterminant()`
and `estimateAntennaPatternConditionNumber()`.
- Now falls back to F=2 (2F=4) if Ddinv=0,
also equivalent to `lalpulsar`.
- Added unit tests.
- Improved CUDA device info logging.
- For developers: updated coding style to `black` 23.1.0 rules
(mostly newlines policy).