fix(tools): route quick/pro tiers to Haiku/Sonnet on Bedrock

Tier engine strings (claude_haiku/claude_sonnet) were stripped back to
azure_mini by per-method whitelists, so both tiers ran gpt-4o-mini and Pro
charged 2x for the same model. Add a shared DbnBedrockModelRouter::
deploymentForEngine() helper and route the cloud path through it across
summarize, ask, barnevernet, discrepancy, deep-research, and korrespond.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

includes/DbnBedrockModelRouter.php

@@ -73,6 +73,22 @@ final class DbnBedrockModelRouter
         return $route['model'] ?? self::LITELLM_SONNET;
     }
 
+    /**
+     * Maps a quality-tier engine string to the LiteLLM deployment name passed to
+     * withDeployment(). claude_haiku/claude_sonnet route to Bedrock Claude when the
+     * active gateway is Bedrock; otherwise they degrade to the Azure GPT-4o family.
+     */
+    public static function deploymentForEngine(string $engine, bool $isBedrock): string
+    {
+        switch ($engine) {
+            case 'claude_sonnet': return $isBedrock ? self::LITELLM_SONNET : 'gpt-4o';
+            case 'claude_haiku':  return $isBedrock ? self::LITELLM_HAIKU  : 'gpt-4o-mini';
+            case 'azure_full':    return 'gpt-4o';
+            case 'azure_mini':
+            default:              return 'gpt-4o-mini';
+        }
+    }
+
     public static function supportsThinking(string $modelName): bool
     {
         return in_array($modelName, self::THINKING_MODELS, true);