Make locked HTTP audit stream IDs server-owned

[aruintelligence]

Problem

Under locked_policy, HTTP deployment evaluation rejected caller changes to profile, mode, failure mode, context, and timestamp, but accepted stream_id. A request with stream_id: "caller-controlled-audit-id" returned policy_source: "server" and embedded that caller value in its runtime audit stream. That weakened the meaning of the locked audit provenance.

Change

• Reject stream_id as a caller control under locked policy, including explicit null, for single and batch endpoints.
• Assign each locked single evaluation a fresh server-generated UUID. Separate evaluations of identical intent have distinct audit streams; their receipt hashes differ by design.
• Preserve caller-selected IDs for the deliberately unlocked demonstration path.
• Update the locked deployment guide and OpenAPI description.

The UUID identifies one server evaluation. It does not authenticate the caller, prove truth of an intent, or claim independent validation. The general deterministic local proof still uses caller-fixed inputs.

Verification

• Reproduced the old behavior: locked HTTP returned status 200 and embedded a caller-supplied audit stream ID.
• node --test test/locked-policy.test.js test/http-server.test.js — 26 passed.
• node --test --test-reporter=dot — passed.
• node scripts/check-public-contracts.mjs, node scripts/check-deployment-surfaces.mjs, npm run check:links, git diff --check — passed.