Don't set dataset peek for errored jobs

This is consistent with the extended metadata logic, and prevents:
```
Message
Job wrapper finish method failed
Stack Trace

Newest

IndexError
list index out of range
```
which only makes it harder to diagnose the actual error, which in the
case of https://sentry.galaxyproject.org/share/issue/831261683ce642a4976efa20a9a0e55c/
is `\nslurmstepd: error: *** JOB 1910486 ON js2-gpu-small0 CANCELLED AT 2024-05-26T23:30:00 DUE TO TIME LIMIT ***\n`

lib/galaxy/jobs/__init__.py

@@ -1786,13 +1786,14 @@ def _finish_dataset(self, output_name, dataset, job, context, final_job_state, r
                     working_directory=self.working_directory,
                     remote_metadata_directory=remote_metadata_directory,
                 )
-            line_count = context.get("line_count", None)
-            try:
-                # Certain datatype's set_peek methods contain a line_count argument
-                dataset.set_peek(line_count=line_count)
-            except TypeError:
-                # ... and others don't
-                dataset.set_peek()
+            if final_job_state != job.states.ERROR:
+                line_count = context.get("line_count", None)
+                try:
+                    # Certain datatype's set_peek methods contain a line_count argument
+                    dataset.set_peek(line_count=line_count)
+                except TypeError:
+                    # ... and others don't
+                    dataset.set_peek()
         else:
             # Handle purged datasets.
             dataset.blurb = "empty"