Describe the splits as the OpenVLHarness evaluation datasets

Drop the *_test_mini splits; mini-vlm-toolkit samples the 300-question default itself

Add hle, MMSearchPlus, MMBrowseComp and VisBrowseBench

Add general-VQA and spatial benchmarks (RealWorldQA, HallusionBench, V*, FREAK, MMSI, ERQA, SPAR, MindCube, EmbSpatial, OmniSpatial) (#3)

Add MMSearch, BrowseCompVL, RealXBench, VDRBench, SimpleVQA(_EN) and WorldVQA_VDR300

README: split into counting & grounding and multimodal search sections

Add FVQA, InfoSeek and EncyclopedicVQA evaluation splits (full + _mini) (#2)

Backup of processed counting/grounding splits (same as bryanzhou008/mini-vlm-toolkit-data)

initial commit