Add Continue for failed local jobs and show project/run names

[bghira]

Failed local jobs now offer Continue, which selects their saved configuration and invokes the existing Run flow with its configured checkpoint resume setting. Configuration activation or loading failures stop the action before training starts. Click and keyboard events stay within the Continue button.

Job cards and search now use project / run names, including tracker metadata already stored in local job history. Cloud submissions persist project identity alongside the run name.

Closes #3204.

Validation:

• Red regressions: Continue absent in real Chrome; four project/run identity tests failed; upload metadata omitted the project for both supported config key styles.
• npm test -- --runInBand tests/js/cloud_utilities.test.js tests/js/cloud_jobs.test.js tests/js/job_submission.test.js: 133 passed.
• .venv/bin/python -m unittest -v tests.test_cloud_job_tracker_metadata tests.test_training_service tests.test_training_service_gpu: 63 passed.
• SIMPLETUNER_SELENIUM_TESTS=1 .venv/bin/python -m unittest -v tests.test_webui_e2e.FailedLocalJobContinueTestCase tests.test_webui_e2e.EasyModeFormDirtyTestCase.test_easy_mode_select_enables_save: 4 passed in Chrome. Covers the saved checkpoint request, activation failure, config loading failure, and click/keyboard propagation.
• Additional form dirty lifecycle, direct Easy Mode/main form edits, and Save As checks passed. The existing Easy Mode select check failed once in the broader run, then passed on both baseline and fixed code in focused runs.

Browser tests inspect the existing Run request without launching model training. Continue preserves the saved resume setting; it does not add checkpoint compatibility or recovery behavior.