fix: prevent evaluation run from getting stuck in Running state when interrupted

[octo-patch]

Fixes #436

Problem

When an evaluation run is interrupted (e.g. via KeyboardInterrupt / Ctrl-C) or
raises an unexpected exception, write_evaluation_run_end was never called.
The run would remain permanently shown as Running in ell-studio, and no new
runs could be added to the same logdir without creating a fresh one.

The TODO comment in the original code acknowledged this was missing:

return evaluation_run
# TODO: add error handling and unsccessful runs.

Solution

• Add a BaseException handler (catches both regular exceptions and
  KeyboardInterrupt) that sets success=False, records the error string,
  and calls write_evaluation_run_end before re-raising. This ensures the run
  is always finalized in the store, regardless of how execution exits.
• Guard EvaluationResults.from_rowar_results against an empty rowar_results
  list, which would previously raise an IndexError when the run is interrupted
  before any results are collected.

Testing

Manually verified by running an evaluation and pressing Ctrl-C mid-run:

• Before the fix: the run stays as Running in ell-studio forever.
• After the fix: the run is immediately marked as failed with the interruption error.