Description
Currently it is apparently required to provide an exactly sized array to the tensor static factory methods.
In the case where (for example) tensors are being created with variable batch sizes, the size of the input will vary. With the current design, it is necessary to allocate a new array appropriate for each batch (and one cannot use a pooled object which may be oversized).
Instead, callers may prefer to allocate a large fixed managed array and then only pass the appropriately sized Span over this array to the factor method, thereby saving an object allocation on every call.
Example given below of the suggested API enhancement (to be replicated across the various data types)
EXISTING: public static Tensor tensor(short[] rawArray, ...
DESIRED: public static Tensor tensor(ReadOnlySpan<Span> rawSpan, ...