Improve Deep RL certification Space runtime pins and writeback errors

For an issue observed in this thread: https://huggingface.co/static-proxy/discuss.huggingface.co/t/error-generating-deep-rl-course-certificate/177241

---

## Summary

This PR makes the Deep RL Course Certification Space easier to rebuild and improves the error message around the certificate writeback step.

Changes:
- Pin Python to 3.10 because a fresh build currently uses Python 3.13 and `Pillow == 9.4.0` fails to build there.
- Pin `huggingface_hub<1` as a temporary compatibility fix because the Space imports the legacy `Repository` class.
- Add input validation for empty username / name fields to avoid accidentally scanning all matching public Hub models.
- Replace the generic Gradio failure around the certified-users writeback step with a clearer message and a logged exception.

This PR intentionally does **not** return a certificate when the writeback step fails, because `certified_users.csv` may be intended as the official issuance record. It only makes the failure explicit instead of surfacing a generic `Error`.

## Investigation notes

I duplicated the Space and tested the current code path with the affected username `szymon-hyziak`.

Observed:
- With Python 3.10, `huggingface_hub<1`, and a valid `HF_TOKEN`, the UI starts.
- The app finds the user's models, downloads/parses the model card metadata, and computes `pass_percentage ~= 81.8`, which should qualify for a completion certificate.
- If the certified-users writeback step is bypassed, the certificate image and progress table render correctly.
- In my duplicate, the remaining failure is the final push to `huggingface-projects/Deep-RL-Course-Certification`, which my token understandably cannot write to.

## Maintainer-side follow-up required

This PR cannot update or verify the official Space secret.

Please check:
- whether the official Space `HF_TOKEN` secret is still valid;
- whether the token owner still has write access to `huggingface-projects/Deep-RL-Course-Certification`;
- whether the official Space logs show a failure around `repo.push_to_hub(...)`;
- whether the Space should be restarted/rebuilt after updating the secret.

I cannot prove the official Space fails at exactly the same line because I do not have access to the official Space logs or maintainer token, but the duplicated Space strongly points to the certificate writeback path rather than the user's progress/model metadata.

README.md

@@ -5,6 +5,7 @@ colorFrom: red
 colorTo: yellow
 sdk: gradio
 sdk_version: 5.36.2
+python_version: "3.10"
 app_file: app.py
 pinned: false
 ---

app.py

@@ -7,12 +7,15 @@ from PIL import Image, ImageDraw, ImageFont
 from datetime import date
 import time  
 
+import logging
 import os
+import requests
 import pandas as pd
 
 from utils import *
 
 api = HfApi()
+logger = logging.getLogger(__name__)
 
 DATASET_REPO_URL = "https://huggingface.co/datasets/huggingface-projects/Deep-RL-Course-Certification"
 CERTIFIED_USERS_FILENAME = "certified_users.csv"
@@ -140,6 +143,15 @@ def check_if_passed(model):
 
 
 def certification(hf_username, first_name, last_name):
+  hf_username = (hf_username or "").strip()
+  first_name = (first_name or "").strip()
+  last_name = (last_name or "").strip()
+
+  if not hf_username or not first_name or not last_name:
+    message = "Please fill in your Hugging Face username, first name, and last name before checking certification."
+    certificate = Image.new("RGB", (100, 100), (255, 255, 255))
+    return message, "./fail.pdf", certificate, pd.DataFrame(), output_row.update(visible=False)
+
   results_certification = [
       {
           "unit": "Unit 1",
@@ -283,6 +295,24 @@ def certification(hf_username, first_name, last_name):
   
   return message, pdf, certificate, df, output_row.update(visible=visible) 
 
+def try_add_certified_user(hf_username, first_name, last_name, pass_percentage):
+  try:
+    add_certified_user(hf_username, first_name, last_name, pass_percentage)
+    return True
+  except Exception:
+    logger.exception("Failed to update certified users dataset")
+    return False
+
+def writeback_error_response():
+  certificate = Image.new("RGB", (100, 100), (255, 255, 255))
+  pdf = "./fail.pdf"
+  message = """
+  Your progress appears to qualify for a certificate, but an internal error occurred while saving your result.
+
+  Please contact the course maintainers and include your Hugging Face username and a screenshot of this page.
+  """
+  return certificate, message, pdf, False
+
 """
 Verify that the user pass.
 If yes:
@@ -312,7 +342,8 @@ def verify_certification(df, hf_username, first_name, last_name):
     certificate, pdf = generate_certificate("./certificate_models/certificate-excellence.png", first_name, last_name)
 
     # Add this user to our database
-    add_certified_user(hf_username, first_name, last_name, pass_percentage)
+    if not try_add_certified_user(hf_username, first_name, last_name, pass_percentage):
+      return writeback_error_response()
     
     # Add a message
     message = """
@@ -328,7 +359,8 @@ def verify_certification(df, hf_username, first_name, last_name):
     certificate, pdf = generate_certificate("./certificate_models/certificate-completion.png", first_name, last_name)
 
     # Add this user to our database
-    add_certified_user(hf_username, first_name, last_name, pass_percentage)
+    if not try_add_certified_user(hf_username, first_name, last_name, pass_percentage):
+      return writeback_error_response()
 
     # Add a message
     message = """

requirements.txt

@@ -1 +1,2 @@
-Pillow == 9.4.0
+Pillow == 9.4.0
+huggingface_hub<1