gdalwarp: add arbitrary percentile resampling (Pxx)

[sidhansu10]

Summary

This PR adds support for arbitrary percentile resampling to gdalwarp
and the warp kernel.

Users can now specify resampling methods such as -r P95, -r P50, etc.,
generalizing the existing quartile and median logic (q1, med, q3).

Initial support is limited to the warp code path (gdalwarp / raster
reprojection). Resize and overview support can be added in follow-up work.

Details

• Introduce GRA_Percentile resampling algorithm
• Extend GDALGetWarpResampleAlg() to parse Pxx values
• Reuse existing quantile logic in GWKAverageOrModeThread()
• Percentile value is validated (0 < P < 100)
• Add CLI regression tests:
  • P50 ≡ med, P25 ≡ q1, P75 ≡ q3
  • Robustness to strong outliers (P95)
  • Invalid values (P0, P100)

Tests

Added tests to autotest/utilities/test_gdalwarp.py.

All existing tests pass.

Notes

Design discussion referenced: warp-only support first, with possible
extension to resize / overviews later.

[rouault]

We shouldn't do that way. Config options are global and messy to deal with. Ideally we would have some C structure describing more about resampling methods than just their identifier., but that would be a quite substantial change. The probably more reasonable way is to request than when GRA_Percentile is set a warping options passed through GDALWarpOperation::papszWarpOptions is specified

[sidhansu10]

Thanks for the review. I’ve updated the implementation to pass the percentile through papszWarpOptions (using a PERCENTILE warping option) instead of a global config option, and updated the warp kernel accordingly.

[rouault]

This call to CPLSetConfigOption() should still be removed. I believe this function should likely take a CPLStringList& aosOptions output argument, and you would do aosOptions.SetNameValue("PERCENTILE", CPLString().Printf("%f", dfPercentile)). And you should "connect the dots" somewhere by actually adding the options filled by GDALGetWarpResampleAlg() to the papszWarpOptions used by gdalwarp_lib.cpp