[Feature Request] Add Quantized Matmul Work-Area Size Helper API

[k8ika0s]

Proposed PR branch: qe/issue-10-quantized-matmul-work-area-size

Category: Feature Request (performance + usability)

Problem

• Quantized matmul can allocate internal scratch when work_area == NULL.
• Callers need a sizing helper for deterministic preallocation.

Justification / why this is needed

• Quantized matmul is typically used in hot paths; allocator overhead here is
  especially costly.
• Sizing helper allows frameworks to allocate once, reuse, and avoid spikes.

Proposed change

• Add zdnn_get_quantized_matmul_work_area_size(...).
• Add tests.
• (Optional follow-up) Add README guidance for caller preallocation in quantized matmul hot paths.

Acceptance criteria

• Helper computes required bytes (return 0 for invalid inputs).
• Tests cover edge cases and representative shapes.

Test plan

• make test (expects tests/testDriver_quantized_matmul_work_area_size_apis.c to pass).

References

• Key files: zdnn/quantized_matmul_work_area.c, zdnn/zdnn.h, zdnn/zdnn.map,
  tests/testDriver_quantized_matmul_work_area_size_apis.c.