Pyfstat

- GridSearch: will now always include the end point in each 1D parameter points array
- restored MCMCSearch.plot_chainconsumer() to a workable state
- optional dependency checked and documented more cleanly
- the injections parameter option for this is actually called "truth" not "truths
- fixed checks of injection parameter keys for other MCMC plotting functions

- improved GridSearch.check_old_data_is_okay_to_use()
- minor logging improvements for Writer, GridSearch, TransientGridSearch
- new binary_mcmc_vs_grid example
- minor improvements to tests and other examples

- LALSuite 6.76 now required
- predict_fstat() helper function made more flexible
- Writer: fixed check_cached_data_okay_to_use() for multiple IFOs
- MCMCSearch and derived classes now store sampler as an attribute
- fixes to MCMC plotting
- proper scaling of injection parameters in walker plots
- proper display of prior (not logprior) in prior_posterior comparisons
- consistently close figure objects after saving
- all classes now announce their creation to the logger
- flake8 compliance
- code style now complying to stricter black 20.8b standards
- improvements to examples

- Writer: fix binary tp handling and clean up injection parameter parsing
- MCMCSearch plotting improvements:
- add injection parameters through "truths" kwarg
- F->2F fixes
- fix failures for single-parameter posteriors
- removed unused Bunch class
- refactored some core functions into BaseSearchClass methods
or moved to helper_functions
- removed deprecated options/functions:
- ComputeFstat option estimate_covering_band
- Writer options minStartTime, maxStartTime
- MCMCSearch.get_median_stds()
- new internal class SearchForSignalWithJump
which SemiCoherentGlitchSearch and GlitchWriter inherit from

- Writer classes, including several backwards-incompatbile changes:
- removed lots of default values where there isn't really
a logical "default"; e.g. detectors and signal parameters
now always need to be explicitly set.
- more flexible setup getting info from noiseSFTs
(tstart, duration now optional and used as constraints)
- can now auto-estimate SFT frequency band if not set by user
(assuming a single-template search)
- added manual ephemerides options
- removed add_noise option, same behaviour can still
be controlled through options sqrtSX or noiseSFTs
- no .cff file generated any more if h0==0 (no signal)
- verbose option for make_cff() and make_data() methods
- clearer error messages in many cases
- ComputeFstat, SemiCoherentSearch and derived classes:
- fixed internal maxStartTime default if not set by user
- by that, fixed the SemiCoherentSearch segments auto-setup
- added Tsft as user option
- clearer error and logging messages
- improved helper_functions.get_sft_array()
- extended, cleaned up and further modularised test suite
- updated examples to changes in Writer and other classes

- fixed semi-coherent search bug introduced in 1.5.0:
- last segment was always skipped due to off-by-one error
in segment boundary calculation.
- MCMC searches:
- extended print_summary()
with new get_summary_stats() helper
and deprecated get_median_stds()
- fixes to some of the more exotic prior types.
- Extended MCMC test coverage.