refactor(open-sse): registry consolidation + DRY media/oauth/adhoc cleanup

- Single-source registry: oauth clientId/tokenUrl, usage URLs, image/embed
  configs, search defaultModel, codex fixedPort, google token url derive.
- Remove 29 unused OmniRoute providers (registry 100→71); media intact.
- De-adhoc: codex literals → registry format/oauth flags; reasoningInject,
  image/embed openrouter headers + xai bodyFields config-driven.
- Add REGISTRY_TEMPLATE.js + expand PROVIDER_DEFAULTS/schema JSDoc.
- Baselines updated; PROVIDERS 62 + alias 90 byte-for-byte, golden snapshots.

Co-authored-by: Cursor <cursoragent@cursor.com>

open-sse/utils/reasoningContentInjector.js

@@ -1,15 +1,12 @@
 // Some thinking-mode providers (DeepSeek, Kimi, MiniMax, ...) require reasoning_content
 // to be echoed back on assistant messages. Clients in OpenAI format don't send it,
 // so we inject a non-empty placeholder to satisfy upstream validation.
+import { PROVIDERS } from "../config/providers.js";
 
 const PLACEHOLDER = " ";
 
-// Provider-level rules: keyed by executor.provider
-const PROVIDER_RULES = {
-  deepseek: { scope: "all" },
-  minimax: { scope: "all" },
-  "minimax-cn": { scope: "all" }
-};
+// Provider-level rules derive from registry transport.reasoningInject (single source)
+const providerRuleFor = (provider) => PROVIDERS[provider]?.reasoningInject;
 
 // Model-level rules: matched by predicate against model id
 const MODEL_RULES = [
@@ -71,7 +68,7 @@ function applyDeepSeekV4ProAlias({ provider, model, body }) {
 }
 
 export function injectReasoningContent({ provider, model, body }) {
-  const providerRule = PROVIDER_RULES[provider];
+  const providerRule = providerRuleFor(provider);
   const modelRule = MODEL_RULES.find(r => r.match(model));
   const rule = providerRule || modelRule;
   const nextBody = applyDeepSeekV4ProAlias({ provider, model, body });