feat(capi): bind task-local training losses

[GBeurier]

Summary

• add a canonical core API that binds a phase-filtered training-loss role to its exact NodeTask attestation
• expose task-bound local loss invocation through the C ABI for downstream R and MATLAB scheduler bridges
• reuse the core binding from WASM and extend generated C fixtures plus negative lifecycle coverage
• refresh ABI, criteria, training, replay, and tracked PyO3 artifacts

Validation

• cargo test -p dag-ml-core -p dag-ml-capi -p dag-ml-wasm (755 passed, 2 ignored)
• cargo clippy -p dag-ml-core -p dag-ml-capi -p dag-ml-wasm --all-targets -- -D warnings
• python3.11 scripts/validate_contracts.py --require-sibling --sibling-root /home/delete/nirs4all/dag-ml-data
• python3.11 scripts/validate_training_replay_contracts.py
• python3.11 -m pytest parity/training/tests/test_training_replay_contracts.py -q (20 passed)
• python3.11 -m pytest parity/criteria/tests/test_criteria_contracts.py -q (10 passed)
• python3.11 scripts/validate_abi_snapshot.py
• python3.11 scripts/validate_release_metadata.py
• python3.11 scripts/check_so_freshness.py
• PYTHONPATH=crates/dag-ml-py/python python3.11 scripts/smoke_python_bindings.py

Review

Independent read-only review completed; final re-review reported no findings.