fix(vision): detect vision-capable custom providers via ProviderProfile flag

_supports_media_in_tool_results() had a hardcoded provider allowlist
that missed custom providers and newer vision-capable providers like
xiaomi. Added ProviderProfile.supports_vision flag and made the
function check:

1. Registered provider profile (supports_vision flag)
2. Model capabilities from models.dev catalog (supports_vision)
3. Existing hardcoded allowlist (unchanged)

This fixes HTTP 400 "text is not set" errors when vision-capable
custom providers receive text-only tool results instead of
multipart image content.

Related: #25594

providers/base.py

@@ -56,6 +56,15 @@ class ProviderProfile:
     auth_type: str = "api_key"   # api_key|oauth_device_code|oauth_external|copilot|aws_sdk
     supports_health_check: bool = True  # False → doctor skips /models probe for this provider
 
+    # ── Vision support ────────────────────────────────────────
+    # True when the provider's API accepts image content inside
+    # tool-result messages natively.  Set on providers that expose
+    # multimodal models via tool results (Anthropic Messages API,
+    # OpenAI Chat Completions, Gemini, Xiaomi, MiniMax, etc.).
+    # Falls back to model-catalog lookup when False and the provider
+    # has no registered profile.
+    supports_vision: bool = False
+
     # ── Model catalog ─────────────────────────────────────────
     # fallback_models: curated list shown in /model picker when live fetch fails.
     # Only agentic models that support tool calling should appear here.