Add repair_run action to manage_job_runs MCP tool

[jacksandom]

Summary

• Adds repair action to manage_job_runs MCP tool, enabling retry of only failed tasks instead of re-running entire jobs
• Implements repair_run() core function following the existing run_job_now pattern
• Supports rerun_all_failed_tasks, rerun_dependent_tasks, rerun_tasks, and latest_repair_id

Closes #392

Problem

When a job run had failed tasks, the LLM had no way to repair the run via MCP. The repair action was missing from manage_job_runs, so attempts fell through to a ValueError. The LLM then fell back to run_now, re-running all tasks (including successful ones), wasting compute and time.

Changes

File	Change
databricks-tools-core/.../jobs/runs.py	New repair_run() function
databricks-tools-core/.../jobs/__init__.py	Export repair_run
databricks-mcp-server/.../tools/jobs.py	Import, 4 new params, "repair" action dispatch with error handling
databricks-tools-core/tests/.../conftest.py	failing_notebook_path fixture
databricks-tools-core/tests/.../test_runs.py	TestRepairRun class

Test plan

• All 11 integration tests pass (including new TestRepairRun)
• MCP smoke test: repair with rerun_all_failed_tasks=True returns repair_id
• MCP smoke test: chained repair with latest_repair_id
• Existing tests unaffected