Gridpp

Force numpy version to be 1.x

New features:
- Improved computational performance of optimal interpolation functions by around 10%
- Added `mask_threshold_downscale_quantile` to compute quantiles and downscale in one go (thanks LonnekevB)
- Added points-to-points gridding methods
- Added vector version of gridpp::interpolate
- Added `RandomChoice` statistic. For functions that compute statistics on vectors, this allows you to randomly pick an element in the vector.
- Added `neighbourhood_score` function that computes a verification score for all observations in a neighbourhood
- Added `gridding` and `gridding_nearest` methods that takes points as inputs and points as output
- Better handling of repeated x-values in `apply_curve`
- Raise exception if constructing `Grid` or `Points` objects with undefined (e.g. NaN) coordinates
- Pre-compiled linux packages for Python 3.7 to 3.11 available

Bug fixes:
- Fixed `bilinear` interpolation for unusual cases (polar stereographic case covering the north pole; thanks trygveasp)
- Fixed bug in structure functions across dateline (thanks joewkr)
- Fixed `optional_interpolation_full` for points (thanks Christoph Herbert and Ewan Pinnington)

Fixed bug in command-line client when using log transform in -d gradient

- optimal_interpolation_full that returns analysis variance
- Point-to-point variant of nearest neighbour and simple gradient downscaling
- allow_extrapolation option in OI functions
- local_distribution_correction method
- doping_square and doping_circle methods
- neighbourhood_search function
- allow_extrapolation boolean to optional_interpolation function
- full_gradient to downscale with spatially varying gradients
- calc_gradient to compute neighbourhood gradient
- grid-to-grid and point-to-point variants of distance and count functions
- sea level pressure function
- window function to compute moving window statistics for timeseries
- Gamma transformation
- Calibration with spatially varying calibration curve
- Speed improvement and bux fixes in bilinear function
- Faster implementation of nearest neighbour interpolation for multiple timesteps
- New way of specifying curve-based correction

* Bug fix to ensemble version of neighbourhood_quantile_fast for quantiles that are varying in space

* Bug fix to neighbourhood_quantile_fast for quantiles that are varying in space