Fix resume after retry failing with version_mismatch error

[neoneye]

Summary

• The frontend /plan/retry endpoint did not set pipeline_version in task.parameters, unlike the MCP cloud equivalent (mcp_cloud/db_queries.py:199)
• This caused the flow Retry → Stop → Resume to fail with "Resume unavailable — plan was created with a different PlanExe version" because pipeline_version was None in parameters
• Fix: set pipeline_version = PIPELINE_VERSION in both /plan/retry and plan creation (/run) to match MCP behavior

Root cause

The frontend had an asymmetry with the MCP cloud endpoints:

• mcp_cloud plan_create / plan_retry / plan_resume all set pipeline_version in parameters
• Frontend /run and /plan/retry did not set it, so old plans and retried plans had no pipeline_version
• Frontend /plan/resume checks parameters.pipeline_version against PIPELINE_VERSION — found None != 1 → version mismatch

Test plan

• Take an old failed plan without pipeline_version in parameters
• Click Retry → let it process a few steps → click Stop
• Click Resume → verify it starts resuming (no version_mismatch error)
• Create a new plan → stop it → verify Resume works
• Verify the pipeline_stop_requested.txt is cleaned up by the worker during resume (existing behavior, line 1199)

🤖 Generated with Claude Code