fix: ask for repo path before building env config in sweagent sh - #1603

Reference Issues/PRs

No existing issue or PR covers this (searched the tracker for sweagent sh / repo prompt / run_shell; the last commit touching sweagent/run/run_shell.py is an unrelated typo fix).

What does this implement/fix? Explain your changes.

Summary

sweagent sh prompts for the repository path when -r/--repo is not given, but the prompt ran after the environment config had already been built. Without -r the docker volume mount therefore became -v None:/repo, and the path entered at the prompt was discarded. This PR moves the prompt above the environment-config construction so the entered path is used.

Why

• sweagent/run/run_shell.py:101 defines -r/--repo with default=None.
• run_from_cli interpolated cli_args.repo into the volume mount while building EnvironmentConfig (docker_args=["-v", f"{cli_args.repo}:/repo"]).
• The if cli_args.repo is None: cli_args.repo = Path(Prompt.ask(...)) block that fills the value in interactive mode sat below the EnvironmentConfig/SWEEnv.from_config calls, so without -r the mount was -v None:/repo and the prompted path never reached the environment.
• The sibling -p problem-statement prompt is already placed before its first use; the repo prompt now follows the same ordering.

Changes

• sweagent/run/run_shell.py: moved the cli_args.repo is None prompt block above the EnvironmentConfig construction so the prompted path is interpolated into docker_args. docker_args is the only other use of cli_args.repo in the file.
• tests/test_run_shell.py (new): regression tests asserting that the repo path from the interactive prompt (and from -r) ends up in the docker args as ["-v", "<repo>:/repo"].

Testing

• New tests on unpatched main: pytest tests/test_run_shell.py -v → AssertionError: assert ['-v', 'None:/repo'] == ['-v', '<tmp>/myrepo:/repo'] (1 failed, 1 passed).
• New tests on this branch: pytest tests/test_run_shell.py -v → 2 passed.
• pytest -m "not slow" on this branch: 99 passed, 2 xfailed, plus 12 pre-existing environment failures (missing sweagent console script on PATH, Docker daemon not running, GitHub API rate limiting) that fail identically on unpatched main; the only run-to-run difference is the new regression test failing there.
• pre-commit run --files sweagent/run/run_shell.py tests/test_run_shell.py → all hooks pass (ruff, ruff format, typos, trailing whitespace, check-ast, ...).