fix(hf-space): trust_remote_code for Phi-4-mini + verbose error reporting

User reported live: ZeroGPU call surfaced "(Error)" with no actionable
detail. Two fixes:

1. Phi-4-mini-instruct has the `custom_code` tag on HuggingFace Hub —
it ships a custom Phi3-based modeling.py inside the repo. Loading
it via AutoModelForCausalLM.from_pretrained() without
`trust_remote_code=True` raises a generic Error from the
transformers Auto class. Added the flag to both AutoTokenizer and
AutoModel calls.

2. Bumped ZEROGPU_DURATION_SECONDS default from 120 to 300 (the
Pro-tier max) to give Phi-4-mini's first-call cold start (download
~8GB + load to GPU + first inference) breathing room. Subsequent
warm calls use a fraction of that.

3. Error messages now include str(e) (truncated to 400 chars) alongside
the class name. Format went from:
"failed (Error). Try again..."
to:
"failed.
**OSError:** Can't load tokenizer for 'foo/bar'. (...)
Try again..."
Future failures will be diagnosable from the UI without needing
Space log access.

Tests: 31/31 still pass (no contract changes).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

app.py

@@ -189,7 +189,7 @@ ROOT = Path(__file__).parent
 ANTHROPIC_MODEL_ID = os.environ.get("MODEL_ID", "claude-opus-4-7")
 HF_MODEL_ID = os.environ.get("HF_MODEL_ID", "google/gemma-2-9b-it")
 ZEROGPU_MODEL_ID = os.environ.get("ZEROGPU_MODEL_ID", "microsoft/Phi-4-mini-instruct")
-ZEROGPU_DURATION_SECONDS = int(os.environ.get("ZEROGPU_DURATION_SECONDS", "120"))
+ZEROGPU_DURATION_SECONDS = int(os.environ.get("ZEROGPU_DURATION_SECONDS", "300"))
 MAX_DESCRIPTION_WORDS = int(os.environ.get("MAX_DESCRIPTION_WORDS", "5000"))
 MIN_DESCRIPTION_WORDS = 200
 
@@ -307,14 +307,22 @@ _zerogpu_tokenizer = None
 
 def _load_zerogpu_model():
     """Load the model + tokenizer once. Called lazily on first request
-    so module import stays fast (the model weights are tens of GB)."""
+    so module import stays fast (the model weights are tens of GB).
+
+    `trust_remote_code=True` is required for models tagged `custom_code`
+    on HuggingFace Hub (e.g. Phi-4-mini-instruct uses a custom Phi3-based
+    config that ships modeling files in the repo)."""
     global _zerogpu_model, _zerogpu_tokenizer
     if _zerogpu_model is not None:
         return
-    _zerogpu_tokenizer = _AutoTokenizer.from_pretrained(ZEROGPU_MODEL_ID)
+    _zerogpu_tokenizer = _AutoTokenizer.from_pretrained(
+        ZEROGPU_MODEL_ID,
+        trust_remote_code=True,
+    )
     _zerogpu_model = _AutoModelForCausalLM.from_pretrained(
         ZEROGPU_MODEL_ID,
         torch_dtype=_torch.bfloat16,
+        trust_remote_code=True,
         device_map="auto",
     )
 
@@ -627,16 +635,26 @@ def diagnose(
         raw = _call_model(SYSTEM_BLOCK, user_prompt, provider)
     except Exception as e:
         # API timeout / rate limit / auth / server / network failure
-        # (Anthropic SDK or huggingface_hub InferenceClient).
+        # (Anthropic SDK, huggingface_hub InferenceClient, or
+        # transformers/torch on the zerogpu path). Include both the
+        # exception class AND its string form so unexpected failures
+        # are diagnosable from the UI without server log access.
         model_label = {
             "anthropic": ANTHROPIC_MODEL_ID,
             "huggingface": HF_MODEL_ID,
             "zerogpu": ZEROGPU_MODEL_ID,
         }.get(provider, provider)
+        detail = str(e).strip() or "(no message)"
+        # Cap the detail so we don't spill multi-paragraph tracebacks
+        # into the UI. 400 chars is enough for a stack-trace summary
+        # without flooding the markdown tab.
+        if len(detail) > 400:
+            detail = detail[:400] + "…"
         return (
-            f"⚠ The diagnostic call to {provider} ({model_label}) failed "
-            f"({type(e).__name__}). Try again in a moment, switch providers in "
-            f"the dropdown, or shorten your description.",
+            f"⚠ The diagnostic call to {provider} ({model_label}) failed.\n\n"
+            f"**{type(e).__name__}:** {detail}\n\n"
+            f"Try again in a moment, switch providers in the dropdown, "
+            f"or shorten your description.",
             "",
         )