Refactoring Tensor

Adding context to Tensor
Adding "active area" to Tensor
Forward declaring TensorImpl in Tensor instead of the another way around
CPU storage is unsigned char instead of void *
todo: memory location for device different than CPU
todo: a Tensor can share a TensorImpl with another Tensor
TensorImpl = storage
Tensor = view on storage

Notes:

• Tensor::getImpl() returns a reference not a pointer, thus client code must replace getImpl()-> by getImpl().
• Tensor.hpp does not includes backend\TensorImpl.hpp but backend\TensorImpl.hpp includes Tensor.hpp. Thus client code needs to include backend\TensorImpl.hpp (with or without Tensor.hpp)

include/aidge/data/Tensor.hpp

@@ -488,6 +488,15 @@ public:
     /// @return Linear index of the first Byte_t of the data at given coordinates
     NbElts_t getIdx(std::vector<Coord_t> const &coordIdx) const noexcept;
 
+    /// @brief Get the logical coordinates of the data at given index, to be used only for
+    /// Python binding
+    /// @param coordIdx coordinates of the desired data
+    /// @note The index is expressed in number of elements
+    /// @return Logical coordinates of the data at given index
+    std::vector<Coord_t>
+    getCoordOnlyForPythonIfYoureUsingItYoureProbablyDoingSomethingWrong(
+        NbElts_t const flatIdx) const noexcept;
+
 private:
     /// @brief Getting the address of the very first data in memory (lexicographic
     /// order), read only access to data