Hailo Wrapper

[m6keller]

• ## Problem

Current hailo-asteroid lacks integrated Hailo-oriented ConvTasNet components and exports, making it harder to use a single consistent API surface for Hailo-targeted model assembly.

Solution

This PR adds Hailo-specific ConvTasNet building blocks and model wrappers, then wires them into package exports.

Added

• asteroid/masknn/hailo_activations.py
• asteroid/masknn/hailo_norms.py
• asteroid/masknn/hailo_convolutional.py
• asteroid/models/hailo_conv_tasnet.py
• asteroid/models/hailo_conv_tasnet_submodules.py

Updated

• asteroid/masknn/init.py
• asteroid/models/init.py

What this enables

• Hailo-specific masknn and ConvTasNet module implementations are now first-class modules in the codebase.
• Public imports are exposed directly from package roots for cleaner downstream usage (e.g. from asteroid.models import HailoConvTasNet).

Validation

• Syntax/bytecode compile checks passed:
  • python -m py_compile asteroid/masknn/hailo_activations.py asteroid/masknn/hailo_convolutional.py asteroid/masknn/hailo_norms.py asteroid/models/hailo_conv_tasnet.py asteroid/models/hailo_conv_tasnet_submodules.py
• Import sanity checks passed:
  • from asteroid.models import HailoConvTasNet
  • from asteroid.masknn import HailoConv1DBlock2D, HailoTDConvNet2D

Impact / Risk

• Scope is additive plus export wiring.
• Main risk is integration behavior in downstream Hailo compilation/runtime pipelines, which is expected and should be covered by follow-up end-to-end validation on target hardware.

Follow-up

• Run Hailo hardware-backed E2E validation (compile/map/runtime latency) using the existing project validation flow.
• Add targeted unit/integration tests for new Hailo modules if desired in a separate PR.

[m6keller]

ooops meant to sent this to my fork!