EMRG v0.5.0

EMRG 0.5.0 adds policy-configurable heuristics while preserving the existing default recommendation behavior.

Highlights

• Added strict version-1 JSON policy support for heuristic thresholds, technique availability, budgets, ZNE factory profiles, scaling methods, PEC samples/noise level, CDR training counts, and composite limits.
• Added optional YAML policy support through emrg[config] using yaml.safe_load().
• Added CLI commands: emrg policy init PATH, emrg policy validate PATH, and emrg generate QASM --policy PATH.
• Added Python API exports: DEFAULT_POLICY, RecipePolicy, load_policy, policy_to_dict, and save_policy.
• Added generate_recipe(..., policy=...), recommend(..., policy=...), and MitigationRecipe.to_dict() for structured recipe serialization.
• Preserved forced-technique warning semantics and clean analyze --json output.

Compatibility

• Default behavior is unchanged when no policy is supplied.
• JSON policy files work with the base install.
• YAML policy files require pip install emrg[config].
• Policy files configure existing EMRG behavior; they do not execute Python or load dynamic imports.

Validation

• ruff check . passed.
• ruff format --check . passed.
• pytest -ra passed: 477 tests.
• pytest --cov=emrg --cov-report=term-missing passed: 96% coverage.
• CLI smoke passed for version, policy init, policy validate, generate with policy, and analyze JSON.
• Benchmarks ran successfully; README benchmark tables were not updated because the fresh run was not uniformly better than the existing snapshot.