[sbrunner]

Summary

When a backport has conflicts, the backport pull request body contains the conflict resolution instructions and a BACKPORT_TODO file is committed to the branch. Once a contributor resolves the conflicts and removes the file, the body kept the stale instructions: the application refuses to overwrite a branch pushed by a human, so create_pull_request never regenerates it. The check action now resets the body once the backport is resolved.

Context

• When the BACKPORT_TODO file is no more present in the backport branch (404 on the content fetch, on the opened, reopened and synchronize events of a ghci/backport/* or backport/* pull request), the body is rewritten to:

  Backport of #<source number> to <target branch>
  
  Conflicts resolved manually
  

Implementation Details

• New _clean_backport_body() helper parsing the head branch name (ghci/backport/<number>-to-<branch>, legacy backport/ prefix also supported) to rebuild the clean message; the format is shared with _backport() through the new _backport_body() helper.
• New _reset_resolved_body() method calling rest.pulls.async_update only when the body was generated by the application (it starts with the clean message and still contains the instructions). A customized body is never touched and the rewrite is idempotent (a clean backport or an already resolved body triggers no API call).

Impact/Risks

• No configuration change, no breaking change: one extra API call on a resolved backport pull request, only while the body still contains the stale instructions.