* load only pycuda engines in pycuda templates

* load only pycuda engines in more templates

* testing idea of skipping gather_dict

* apply error allreduce to other engines

* added new reduced error logic to accelerated stochastic engines

* Fix conda to 24.1.1 because of bug in 24.1.2

* whitespace

* Need to roll back to 23.11.0

* seems like conda 23.11.0 needs python 3.11

* Option to use full polynomial in ML linesearch

* Tidy and add full polynomial for Euclid model

* code restructure, less switching

* change new parameter from boolean to str

---------

Co-authored-by: Benedikt Daurer <benedikt.daurer@diamond.ac.uk>

* WIP kernel implementation of general FFT filter

* Tests for FFT filter

* FFT-based gaussian smoothing in ML

* Add dummy _get_smooth_gradient_fft in ML_serial

* added batched fft_filter tests

* implented numpy based gaussian fft filter and added more tests

* Introduced new parameter for changing method for smoothing kernel

* Fixed smooth gradient method parameter

* add example for FFT smoothing in ML pycuda

* Added new template

* Working on FFT based Gaussian filter for cupy engines

* Fixed another bug in batch multiply kernel

* fft based Gaussian smoothing works with both Ml_pycuda and ML_cupy

* small changes to accelerate tests

* removed debugging traces

* Improve error message when convolution kernel too big

---------

Co-authored-by: Timothy Poon <timothy.poon@diamond.ac.uk>
Co-authored-by: Benedikt Daurer <benedikt.daurer@diamond.ac.uk>

* pump version to 0.8.1

* Updated documentation

* improved installation instructions

* changed filtered cufft instructions to cupy env

* small changes in text

* adding Python 3.12 to workflow
* added more comments in workflow