v4.11.1 — Run to completion on Windows

Summary: The new "keep going until the plan is done" check silently did nothing on Windows whenever the progress notes had been saved with Windows line endings — which a live test run did on its own. It now reads those notes correctly.
Details: A live unattended run on Windows rewrote its scratchpad with Python, which writes CRLF line endings. The status, feature and branch fields then failed to parse, so the run-to-completion check saw no status and let the turn end mid-plan; the session-start state summary lost the same fields. Line endings are now normalized before parsing, with a regression test. Version 4.11.1.
Technical details: One normalization in the shared scratchpad parser, used by both the Stop-hook continuation check and the session-start hook; found by replaying the live run's real transcript against the handler. No screens, endpoints, schema or deployment impact.

Update an installed project: claude plugin update claude-code-sdlc@claude-code-sdlc --scope project