Readiness endpoint can report ready for corrupt runtime artifacts

[Ayushdevo]

Problem

GET /ready currently checks only whether the model and health-report paths exist.

A zero-byte/corrupt baseline_model.joblib or malformed model_health_report.json still makes the endpoint return 200 {"status":"ready"}, while /predict or /health can fail immediately afterward.

Why it matters

For orchestration/readiness probes, file existence is weaker than actual service readiness and can route traffic to an unhealthy instance.

Suggested fix

Validate that:

• the model artifact can be loaded successfully, and
• the health report can be parsed as JSON with the expected top-level fields.

Keep the check lightweight; caching or mtime-based validation is fine if repeated deserialization is a concern.

Acceptance criteria

• corrupt model artifact => /ready returns 503
• malformed health JSON => /ready returns 503
• valid artifacts => /ready returns 200
• regression tests cover all three cases