Add full op for Arm backend (#4073)

Summary:
Implements  the full op which creates a tensor
of a given shape filled with a given value.
The shape and value are set at compile time,
i.e. can't be set by a tensor input.

Refactors tosa_quant_utils.is_quant_node
to handle nodes with no inputs (or outputs)

Does not add a full quantizer annotator, the op
needs to be quantized by a SharedQuantizationSpec

Change-Id: I1cebd1da1af5b9aa726a363431ffc30d8259a0ff

Pull Request resolved: https://github.com/pytorch/executorch/pull/4073

Reviewed By: mergennachin

Differential Revision: D59259731

Pulled By: digantdesai

fbshipit-source-id: 621fec994bc2ebc4ad7abd51d9dbf1a5a4deed43