Persist ForexDataProvider across calls — speeds up repeat /scans 50×

Cold /scan is rate-limited (4 pairs × 3 TFs = 12 credits, free tier 8/min)
so the first call takes ~73s. Without provider reuse, every subsequent
/scan is also ~73s because the cache is rebuilt from scratch.

Reusing the provider on the bot instance preserves the OHLCV cache
between calls, so a /scan within the cache TTLs (M15: 5m, H1: 30m,
D1: 4h) skips the prefetch entirely:

  Cold /scan:  ~73s  (rate-limited prefetch + analyze + Gemini)
  Warm /scan:  ~10s  (cached data, news re-scored)
  Re-warm:     ~1.5s (cached data + cached sentiment)

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

telegram_bot.py

@@ -274,6 +274,10 @@ class ForexTelegramBot:
         self._executor = ThreadPoolExecutor(
             max_workers=len(SCAN_TIMEFRAMES), thread_name_prefix="forex"
         )
+        # Persistent ForexDataProvider so the OHLCV cache survives across
+        # /scan and /signal calls. Without this each call burns API credits
+        # re-fetching the same bars and waits up to 60s on the rate limit.
+        self._provider = ForexDataProvider(PlatformConfig())
 
     # ── Engine helpers ────────────────────────────────────────────────
 
@@ -297,7 +301,7 @@ class ForexTelegramBot:
             for tf_name in SCAN_TIMEFRAMES
             for tf in [_TF_MAP[tf_name]] + _TF_CONFIRMATIONS.get(tf_name, [])
         })
-        provider = ForexDataProvider(PlatformConfig())
+        provider = self._provider  # reuse across calls — see __init__
 
         loop = asyncio.get_running_loop()
         if progress: