Use gpt-6-luna for implementation workers, gpt-6.1-sol for highly complex features

[coneilen]

Summary

Operator decision (2026-10-14): implementation workers use gpt-6-luna, high, by default, and gpt-6.1-sol, high, long_context only for highly complex features. Planning workers and the coordinator stay on gpt-6.1-sol, high (#806). The reviewer stays a different vendor.

Changes

• investigation/windows-parity-orchestration-plan.md:
  • the model table now has "Implementation (default)" on gpt-6-luna and "Implementation (highly complex features)" on gpt-6.1-sol, replacing the per-issue "small fixes" and "lane implementation" rows;
  • wave-2 step 3 names both models.
• investigation/windows-parity-lane-kickoff.md: the planner rates complexity (standard or highly complex, with the reason), and that rating picks the implementer.

Test plan

RED: Select-String origin/main plan for "Implementation (default)" -> 0 matches; lane implementation defaulted to gpt-6.1-sol
GREEN: Select-String edited plan for "Implementation (default)" -> 1 match (gpt-6-luna); kickoff planner step 7 rates complexity
REGRESSION: pwsh Tools/windows/validate.ps1 -Task privacy -> Privacy checks passed; git diff --check -> clean; git ls-files --eol -> plan i/crlf preserved; bash Tools/ci/tests/classify-changes.test.sh -> 59 passed, 0 failed

This is a documentation-only change, so the Windows suites and macOS make test/make check were not run.

Checklist

• I have read the Contributing Guidelines
• I have signed off my commits (git commit -s) per the DCO
• Tests pass locally (make test)
• Code follows the existing style (make check)
• I added the test/contract before the implementation and observed the intended RED failure