Add MobileNetV3 operators to aidge_core

This MR is linked to aidge#299 (closed).

This MR aims to add operators declaration to support MobileNet-v3 Small. The current ONNX model (https://huggingface.co/qualcomm/MobileNet-v3-Small/resolve/main/MobileNet-v3-Small.onnx) uses two operators not yet implemented in Aidge:

• HardSigmoid
• HardSwish

HardSwish - Core Framework Implementation

HardSwish is implemented as a meta-operator defined as Mul(x,HardSigmoid(x)).

Core Framework Files

• HardSwish.cpp (53 lines) - Meta-operator implementation with micro-graph construction
• MetaOperatorDefs.hpp (lines 420-445) - Function declarations for HardSwish_Op and HardSwish
• Test_MetaOperator.cpp (lines 246-273) - Unit tests for HardSwish meta-operator functionality

Meta-Operator Architecture

Implementation (HardSwish.cpp)

• Micro-Graph Construction: Creates Identity -> HardSigmoid -> Mul graph structure
• Graph Connections: input -> hardSigmoid, input -> mul, hardSigmoid -> mul
• Parameter Handling: Supports alpha/beta parameters for HardSigmoid component
• Node Management: Proper input/output ordering with setOrderedInputs and setOrderedOutputs

Function Declarations (MetaOperatorDefs.hpp)

• HardSwish_Op: Meta-operator creation with default parameters (alpha=1/6, beta=0.5)
• HardSwish: Node factory function with parameter customization
• Documentation: Comprehensive API documentation with mathematical definition

Unit Testing (Test_MetaOperator.cpp)

• Graph Structure Validation: Verifies 3-node micro-graph (Identity, HardSigmoid, Mul)
• Input/Output Testing: Validates 1 input, 1 output with proper tensor dimensions
• Functional Testing: Tests with actual tensor data and dimension forwarding

HardSigmoid - Core Framework Implementation

Core Framework Files

• HardSigmoid.hpp (142 lines) - Main operator header with attribute definitions and class interface
• HardSigmoid.cpp (49 lines) - Operator implementation with constructor and factory functions
• pybind_HardSigmoid.cpp (86 lines) - Python bindings with comprehensive documentation
• Test_HardSigmoid_Op.cpp (62 lines) - Unit tests for operator attributes and functionality

Core Operator Architecture

Header Implementation (HardSigmoid.hpp)

• Attribute System: Uses StaticAttributes<HardSigmoidAttr> with Alpha/Beta parameters
• Class Definition: HardSigmoid_Op inherits from OperatorTensorWithImpl<HardSigmoid_Op>
• Type Safety: Template-based implementation with proper constexpr definitions
• API Design: Clean interface with alpha/beta getters and factory function

Source Implementation (HardSigmoid.cpp)

• Constructor: Parameterized constructor with alpha/beta attribute initialization
• Copy Constructor: Proper deep copying of attributes and operator state
• Factory Function: HardSigmoid() creates nodes with default or custom parameters
• Attribute Access: Direct access to alpha/beta parameters via reference

Python Bindings (pybind_HardSigmoid.cpp)

• Class Binding: Complete Python class with attribute access and documentation
• Function Binding: Factory function with parameter defaults and type safety
• Documentation: Comprehensive docstrings with mathematical definitions
• Registration: Automatic operator registration for Python interface

Unit Testing (Test_HardSigmoid_Op.cpp)

• Attribute Testing: Validates default and custom alpha/beta parameter handling
• Functionality Testing: Tests constructor and factory function behavior
• Type Safety: Verifies proper operator creation and type casting

External References

• https://onnx.ai/onnx/operators/onnx__HardSigmoid.html
• https://onnx.ai/onnx/operators/onnx__HardSwish.html