Chess-tuning-tools

------------------

Local tuner
~~~~~~~~~~~

* Emit errors and warnings for common cutechess-cli problems:

- Unknown UCI parameters
- An engine loses on time
- The connection to an engine stalls (usually a crash)
- Any other error captured by cutechess-cli
* Add support for specifying nested parameters.

------------------

Local tuner
~~~~~~~~~~~

* Replace default lengthscale priors by inverse-gamma distributions.
* Add the following command line flags, which allow the user to override the
prior parameters:

- ``--gp-signal-prior-scale`` for the scale of the signal prior.
- ``--gp-noise-prior-scale`` for the scale of the noise prior.
- ``--gp-lengthscale-prior-lb`` for the lower bound of the lengthscale prior.
- ``--gp-lengthscale-prior-ub`` for the upper bound of the lengthscale prior.

------------------

Local tuner
~~~~~~~~~~~

* Add ``--fast-resume`` switch to the tuner, which allows instant resume
functionality from disk (new default).
* Fix the match parser producing incorrect results, when concurrency > 1 is
used for playing matches.

Distributed tuning framework
~~~~~~~~~~~~~~~~~~~~~~~~~~~~
* The distributed tuning framework is no longer deprecated.
* Add ``--run-only-once`` flag to distributed tuning client. If True, it will
terminate after completing one job or immediately if no job is found.
* Add ``--skip-benchmark`` flag to distributed tuning client. If True, it will
skip the calibration of the time control, which involves running a benchmark
for both engines.
* Tuning server of the distributed tuning framework will now also save the
optimizer object.
* Tuning server now also uses the updated pentanomial model including
noise estimation.
* ``warp_inputs`` can now be passed via database to the tuning server.
* Fix the server for distributed tuning not sorting the data by job id causing
the model to be fit with randomly permuted scores.
* Fix the server for distributed tuning trying to compute the current optimum
before a model has been fit.

------------------
* Print user facing scores using the more common Elo scale, instead of negative
downscaled values used internally.
* Internal constants set to improved values.
* Always send ``uci`` first before sending ``setoption`` commands to the engine.

------------------
* Fix incorrectly outputting the variance instead of the standard deviation for
the estimated error around the score estimate.

------------------
* Fix a bug where the model was not informed about the estimated noise variance
of the current match.
* Revert default acquisition function back to ``"mes"``.
* Remove noise from the calculation of the confidence interval of the optimum
value.
* Log cutechess-cli output continuously.
* Add ``"debug_mode"`` parameter which will pass ``-debug`` to cutechess-cli.
* Add support for pondering using ``engineX_ponder``.
* Fix passing boolean UCI options correctly.