fix(routing): KI-154 — empty-profile QA must route to fact_find

Live bug: user said 'to buy a health insurance policy' (clear intent to
buy) and got a generic policy-fact dump citing Health Protector Plus +
ACKO + IRDAI 2024 regulations on an EMPTY profile session. Bot quoted
'Any One Illness' definitions to someone who hasn't even given their name.

Two failures combined:
1. FACT_FIND_TRIGGERS only had 'buy health insurance' (no article).
'to buy a health insurance policy' has 'buy A health insurance' — the
substring match failed, classify_intent fell through to 'qa'.
2. should_route_to_fact_find force-routed only recommendation/comparison
intents to fact_find on empty profile. QA was let through, which is
correct for specific-policy questions ('Does Star Comprehensive cover
IVF?') but wrong for generic intent statements ('to buy a policy').

Fix:
1. Extend FACT_FIND_TRIGGERS with more buying phrasings: 'to buy',
'want to buy', 'looking to buy', 'buy a ', 'buy an ', 'purchase',
'shopping for a', 'get a policy', 'need a plan', 'buy a policy', etc.
2. New rule in should_route_to_fact_find: when profile_is_empty AND
intent=='qa' AND query does NOT name a specific insurer/SKU (via
_names_two_specific_policies OR single-insurer-token check), force
route to fact_find. Empty-profile users never see raw policy facts
before we know who they are.

Verified localhost: 'to buy a health insurance policy' now routes to
fact_find_brain with conversational welcome asking age + dependents.
Specific-policy qa queries that DO name an insurer still get qa.

backend/orchestrator.py

@@ -59,6 +59,14 @@ FACT_FIND_TRIGGERS = (
     "have kids", "have parents", "for my parents", "for my family",
     "my income", "earn ", "salary",
     "first policy", "first time buyer",
+    # KI-154 (2026-05-15) — additional buying-intent phrasings that don't
+    # contain "buy health insurance" as a tight substring. Live bug: user
+    # said "to buy a health insurance policy" — fell through to qa because
+    # the literal "buy health insurance" (no article) didn't match.
+    "to buy", "want to buy", "looking to buy", "buy a ", "buy an ",
+    "purchase", "purchasing", "shopping for a", "get a policy",
+    "get a plan", "need a policy", "need a plan", "buy a policy",
+    "buy a plan", "buy insurance", "buy a policy for",
 )
 
 COMPARISON_KEYWORDS = (
@@ -265,6 +273,24 @@ def should_route_to_fact_find(
         if intent == "comparison" and _names_two_specific_policies(query):
             return False
         return True
+    # KI-154 (2026-05-15) — empty-profile + qa intent: route to fact_find
+    # UNLESS the query names a specific policy / insurer (real policy-fact
+    # lookup) or asks about specific policy mechanics. Without this rule,
+    # broad questions like "to buy a health insurance policy" or "tell me
+    # about health insurance" dumped raw policy facts on users who haven't
+    # even given their name yet.
+    if profile_is_empty and intent == "qa":
+        # If query names a specific insurer/SKU, it's a legitimate policy-fact
+        # lookup (e.g., "Does Star Comprehensive cover IVF?"). Let qa handle.
+        if _names_two_specific_policies(query):
+            return False
+        # Single named policy/insurer is also fine for qa
+        ql = query.lower()
+        if any(tok in ql for tok in _NAMED_INSURER_TOKENS):
+            return False
+        # Otherwise — empty-profile generic intent — route to fact_find so the
+        # bot collects basic context before quoting policy specifics.
+        return True
     return False