[Bug] Fix Operation Availability Mapping + Mapping-Driven Tests

[k8ika0s]

Proposed PR branch: qe/issue-14-operation-availability-mapping

Category: Bug (correctness)

Problem

• Operation availability mapping can drift from capability semantics.
• Conv2D mapping and tests needed correction to avoid false positives/negatives.

Justification / why this is needed

• Incorrect availability mapping forces integrators into suboptimal choices:
  skipping NNPA when it is available, or attempting NNPA when it is not.
• Drift is easy because mapping logic is cross-cutting; conformance tests
  prevent regressions when capability rules evolve.
• Baseline+regtests showed a mapping failure that is fixed in enhanced.

Proposed change

• Correct Conv2D availability mapping in capability logic.
• Add mapping conformance tests.
• Update testDriver_available_apis Conv2D expectation to derive from capability bits.

Acceptance criteria

• Availability results match function/parmblock capability bits.
• Tests validate mapping rules and prevent drift.

Test plan

• make test (expects:
  • tests/testDriver_operation_availability_mapping.c
  • tests/testDriver_available_apis.c)

References

• Baseline report: artifacts/FINAL_BASELINE_ENHANCEMENT_REPORT.md shows
  operation availability mapping regression failing on baseline+regtests and passing in enhanced.
• Key files: zdnn/utils.c, tests above.