enh(run): defer prompt_toolkit import until prompting

[dexhunter]

Summary

• defer the private multiline prompt import until the existing no-task branch needs it
• add a fresh-interpreter regression test for the import boundary
• preserve interactive prompting by patching the helper at its defining module

This follows a review suggestion from PR #749.

Performance

Measured with Python 3.11 in the same isolated environment, using 31 fresh interpreters after 5 warmups:

Revision	Median minisweagent.run.mini import
Current main (e187bcb2)	118.993 ms
This change	54.976 ms

That is a 53.80% reduction (2.16x faster) for fresh module import.

This benchmark measures fresh module import only. CLI help is a correctness gate, not a separately timed metric. It does not measure normal default agent execution, and prompt_toolkit remains a required dependency.

Validation

• new subprocess regression test fails on pristine main and passes with this change
• focused tests: 133 passed
• full suite: 568 passed, 51 skipped
• uvx ruff check --fix .
• uvx ruff format --check .
• pylint minisweagent/ --errors-only
• git diff --check

Autoresearch

Supplementary autoresearch: 20-step public dashboard.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
src/minisweagent/run/mini.py	94.87% <100.00%> (ø)

... and 1 file with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[dexhunter]

The only red check is the docs dependency-install step, before repository code runs. The job selected Python 3.14; resolving the current full extra pulled litellm 1.92.0, whose pyo3-ffi 0.23.5 build supports at most Python 3.13 and exits during wheel construction. This PR changes only the prompt_toolkit import location and does not modify dependencies or workflows. The other six checks pass, modified lines are covered, and the PR is approved. The docs job should pass once that environment uses Python 3.13 or the upstream PyO3 constraint is updated.

[dexhunter]

Rebased onto current main at 5cb8970. The full current check set is now green, including the previously failing docs deploy job; the focused lazy-import tests and approval remain in place. This is merge-ready from my side.

[dexhunter]

Re-verified against today's main (a83fcae), four commits past the approved head:

• the change cherry-picks onto main with no conflict, and the top-level _multiline_prompt import is still at mini.py:15, so it still applies exactly as written;
• the regression test still fails against unmodified main and passes with the change;
• re-measured fresh-interpreter import minisweagent.run.mini on Python 3.11, 31 runs after 5 warmups with the two arms interleaved: 204.6 ms -> 104.2 ms median. The absolute numbers are higher than in the description because this box is busier today; net of a 15.1 ms bare-interpreter floor that is 189.5 ms -> 89.1 ms, about 2.1x, in line with the original measurement.

Nothing outstanding from my side. Happy to rebase onto current main, or to split the regression test out on its own, if either would make it easier.