Discovery: merge %-gainer movers (the wildcards volume-ranking can't see)

Forensic from the source trader's 2026-07-02 recap: his ENTIRE day (+$8.9k) was CX — a thin
reverse-split name that squeezed $3->$6 on a headline (his "wildcard effect": on dead pre-holiday
days the one mover gets everyone's attention). Our system NEVER SAW IT: zero bars, zero discovery
events. Root cause: discovery ranks by ABSOLUTE SHARE VOLUME (most-actives), and a thin %-squeezer
never cracks a volume-ranked top-N. Every other name he called out we saw and judged the same way
he did (DSY passed/backed-out, CWD too cheap, AHMA rejected) — the gap was pure discovery scope.

- alpaca_paper/client.py: get_movers(top) — /v1beta1/screener/stocks/movers top %-gainers
  (verified working with our keys: CLRO +114%, CWD +73% live).
- discover_active_symbols: wildcard gainers are fetched FIRST (warrants/units filtered:
  '.'-suffixed + 5-char W-endings), deduped with most-actives, same price-band filter.
  Additive + fail-open: a movers failure never breaks the existing actives path.

Discovery/ingestion scope ONLY — every trading gate (confirmation window, quality, VWAP, cap)
still applies unchanged. The immediate value is that shadow tracks + the labeler finally get
BARS for wildcard movers, so the next CX is measurable instead of invisible.

Verified live: universe now leads with gainer-sourced names (CLRO, CWD, USDE, YRD, DSY). 16 tests pass.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

research/ingestion/market_data.py

@@ -273,9 +273,24 @@ def discover_active_symbols(
     static symbol list.
     """
     try:
+        # WILDCARD movers first: top %-GAINERS regardless of absolute share volume.
+        # A thin name squeezing on a headline (CX 2026-07-02: $3->$6, the source
+        # trader's entire day) never cracks a volume-ranked top-N — this is the feed
+        # that sees it. Discovery/ingestion only; every trading gate still applies.
+        gainers: list[str] = []
+        try:
+            for g in client.get_movers(top=15):
+                sym = (g.get("symbol") or "").strip()
+                # drop warrants/units/rights noise (BKSY.WS, EVLVW-style)
+                if not sym or "." in sym or (len(sym) >= 5 and sym.endswith("W")):
+                    continue
+                gainers.append(sym)
+        except Exception:  # noqa: BLE001 — movers is additive; actives still work
+            pass
         # Alpaca caps the most-actives `top` at 100; never request more.
         actives = client.get_most_actives(top=min(top * 3, 100), by="volume")
-        symbols = [a["symbol"] for a in actives if a.get("symbol")]
+        symbols = list(dict.fromkeys(
+            gainers + [a["symbol"] for a in actives if a.get("symbol")]))
         if not symbols:
             return []
         trades = client.get_latest_trades(symbols)  # batched internally; price them all