fix(group-chat): full desktop parity for turn timeout (180s base + 20m adaptive cap) and multi-profile REST fallback

- Match desktop GROUP_TURN_TIMEOUT_MS (180s) and GROUP_TURN_HARD_CAP_MS (20m)
- Implement adaptive deadline extension: actively streaming tokens or running tools within 60s keeps the turn slot alive up to 20m, preventing long research/terminal runs from getting discarded
- Update LEASE_STEP_TTL_MS to 180s
- Add explicit `@Query("profile") profile: String?` support to `HermesApiService.getSessionMessages`
- Pass `bot.name` explicitly in `GroupChatViewModel` REST fallback so multi-profile bot sessions do not hit 404 under `ProfileScopeInterceptor`

Co-authored-by: Luna <luna.m57@agentmail.to>

app/src/main/java/com/m57/hermescontrol/data/remote/HermesApiService.kt

@@ -194,6 +194,7 @@ interface HermesApiService {
         // legacy backends that don't know the param — they ignore it.
         @Query("order") order: String? = null,
         @Query("include_compacted") includeCompacted: Boolean? = true,
+        @Query("profile") profile: String? = null,
     ): Response<SessionMessagesResponse>
 
     @GET("api/sessions/stats")