Pin semantic extraction temperature

[ferueda]

Summary

• require an explicit numeric temperature: 0 in the reviewed semantic-v1 route
• send that value through the existing OpenAI-compatible Gateway request without adding seed or other sampling controls
• reject missing, null, nonnumeric, or nonzero temperature before adapter construction
• update the example route, regression fixtures, architecture, roadmap, and active-plan status

Why

Semantic claims are an extraction task, so lower sampling variation is preferable to creative output. Making temperature explicit also prevents provider defaults from changing behavior without changing Noema's recorded route configuration and processing identity. Temperature zero improves repeatability but is not treated as a determinism guarantee.

Impact

Previously copied route files must add numeric "temperature": 0 before Noema will accept them. Existing stored analyses remain inspectable. Because the sanitized route is already part of the semantic processing key, the updated route creates new work once and exact later reruns remain reusable. There are no domain or database changes.

No live request was made in this change. The approved gateway check follow-up will validate the pinned parameter against Vercel and Cerebras using public synthetic input.

Validation

• go test ./internal/adapters/aigateway ./cmd/noema
• make check