feat(cache): add OpenAI prompt_cache_key and Gemini cachedContent support

Provider-specific caching enhancements:

OpenAI:
- Auto-generate prompt_cache_key from message prefix hash
- Key format: omni-{prefix_hash_32chars}
- Preserves client-provided keys (doesn't override)

Gemini:
- Preserve cachedContent ID if provided by client
- Enables explicit Gemini caching for long prompts
- cachedContentTokenCount already tracked in responses

src/lib/promptCache/index.ts

@@ -1 +1 @@
-export { analyzePrefix, shouldInjectCacheControl } from "./prefixAnalyzer";
+export { analyzePrefix, shouldInjectCacheControl, generatePromptCacheKey } from "./prefixAnalyzer";

src/lib/promptCache/prefixAnalyzer.ts

@@ -75,3 +75,11 @@ export function analyzePrefix(messages: Message[]): PrefixAnalysis {
 export function shouldInjectCacheControl(analysis: PrefixAnalysis, minTokens = 1024): boolean {
   return analysis.prefixTokens >= minTokens && analysis.confidence >= 0.7;
 }
+
+export function generatePromptCacheKey(messages: Message[]): string {
+  const analysis = analyzePrefix(messages);
+  if (analysis.prefixHash) {
+    return `omni-${analysis.prefixHash.slice(0, 32)}`;
+  }
+  return "";
+}