fix(ai): adaptive max_tokens per backend (reasoning profiles get 1024+, native llama.cpp stays compact; per-profile max_tokens override); include entry texts in ai_status SSE notify so journal updates live without reload; audit log action filter; links pagination; settings validation & UI improvements

public/audit.html

@@ -16,6 +16,9 @@
     </div>
 
     <div class="actions-row">
+      <select id="auditActionFilter" class="settings-input" style="max-width:280px">
+        <option value="">Все действия</option>
+      </select>
       <input type="search" id="auditFilter" class="settings-input" placeholder="Фильтр по действию..." style="max-width:300px">
       <button class="btn-link" id="auditRefreshBtn" type="button" title="Обновить">Обновить</button>
       <span class="hint" id="auditCount"></span>