Implement prompt execution provenance and observability

[andrewperis]

Parent epic: #4

Summary

Implement the Prompt SDK provenance and observability records required to trace how a prompt execution was produced without logging secrets or unnecessary sensitive/private content.

Scope

• Record prompt ID/version and Prompt Specification version.
• Record hashes/identifiers for rendered prompt and relevant context package inputs as policy permits.
• Record provider/model/adapter identity and normalized generation parameters.
• Record execution timestamps, duration, usage metadata, result identifiers, and status.
• Record validation/lint and structured-output validation outcomes.
• Support correlation/execution IDs for downstream manifest/build orchestration.
• Define redaction and safe-logging rules.
• Keep observability sinks pluggable rather than requiring a specific telemetry vendor.

Deliverables

• Execution provenance schema/model.
• SDK instrumentation hooks.
• Redaction/safe-logging policy.
• Default local/test observer implementation.
• Tests and examples.

Acceptance Criteria

• Every SDK execution can produce a stable execution/provenance record.
• Prompt and context versions/identifiers are traceable.
• Provider/model and relevant parameters are traceable.
• Secrets and prohibited sensitive data are not logged by default.
• Observability can be integrated without a specific vendor dependency.
• Records are suitable for later Comic Manifest/Build Orchestrator provenance.
• Tests verify redaction and metadata completeness.

[andrewperis]

Completed through the merged coordinated PRs: DefinitelySecureStudio/codex#9 and DefinitelySecureStudio/platform#12. Includes review hardening; verification passed with 64 Codex conformance cases and 78 Platform tests.