Enables compiler auto-vectorization for same-dtype BinData vectors by replacing ConstDataView/LittleEndian reads with memcpy-based float loads and switching INT8 accumulation from double to int32_t.

Adds a zero-allocation cross-dtype fast path so mixed-dtype BinData comparisons (e.g., FLOAT32 vs INT8) compute directly on raw bytes instead of heap-allocating std::vector<Value>.