Add --lf/--failed to re-run only previously-failed tests

[mvdbeek]

Summary

The --failed flag has been defined in planemo test since 2015 but was never
wired into the modern execution pipeline — the only code that referenced it
(GalaxyTestCommand.build()) is dead code left over from the old run_tests.sh
approach.

This PR implements the flag properly:

• --failed / --lf: filters test cases before execution to only those
  whose IDs appear as non-success in a previous tool_test_output.json.
  --lf is added as a pytest-compatible alias.
• --failed_json: optional separate path to read the previous results from,
  distinct from --test_output_json. This is needed when the planemo-ci-action
  test loop runs multiple tool groups, each writing their own output JSON, while
  all reading from the same previous chunk-level JSON for filtering.

How the filter works

Test case IDs follow the pattern {tool_id}_{index} (e.g. bwa_mem_0).
StructuredData.failed_ids already returns these from the JSON; the new code
in BaseEngine.test() filters the gathered test cases against that set before
handing them to the engine.

planemo-ci-action changes

A companion PR to planemo-ci-action
adds a retest mode and a previous-run-id input. In retest mode the action:

1. Downloads the Tool test output {chunk} artifact from the specified previous
   run via gh run download
2. Finds the same tool groups for this chunk
3. Runs planemo test --failed --failed_json previous_run_results/tool_test_output.json
   for each group

This lets IUC weekly test maintainers re-run only the failing tests from a
previous scheduled run by triggering a workflow_dispatch with the failed run's
ID, without re-testing everything that already passed.

Usage

# normal run
planemo test tool.xml --test_output_json results.json

# re-run only failures (reads results.json, overwrites it)
planemo test tool.xml --lf --test_output_json results.json

# separate input and output (useful in CI loops)
planemo test tool.xml --lf --failed_json prev.json --test_output_json rerun.json

Test plan

• Unit tests in tests/test_last_failed.py covering:
  • StructuredData.failed_ids includes failure/error/skip, excludes success
  • Without --failed, all test cases pass through unchanged
  • With --failed, only previously-failed cases are submitted to the engine
  • --failed_json takes precedence over --test_output_json as input source
  • Missing JSON raises a clear ClickException
  • All-passed previous run returns empty result immediately (no engine calls)