fix(ab-3238): switch drift integration from merge to rebase

[PolyphonyRequiem]

PR #471 (just merged) shipped AB#3238 using git merge --no-ff origin/<target> as the drift-integration strategy. Per user feedback, the feature PR's Commits tab should show clean linear history — not an extra integration merge commit sitting on top of the feature work. Switching to git rebase + git push --force-with-lease.

Why rebase (vs merge)

• Feature PRs are squash-merged into main (per cloudvault config), so preserving upstream MG-merge SHAs on the feature branch adds nothing — they're discarded at squash anyway.
• Rebase keeps the PR's Commits tab focused on the work the feature actually did; reviewer sees upstream drift as ancestor history, not as a sibling merge commit.
• Force-with-lease (refusing if origin advanced since our fetch) + the existing feature_branch_diverged gate make force-push safe given polyphony's branch model (SDLC run is sole writer of the feature branch).

What changed

Script (integrate-target-drift.ps1):

• git merge --no-ff → git rebase origin/<target>
• git push origin <feature> → git push --force-with-lease=refs/heads/<feature>:<expected-sha> origin HEAD:refs/heads/<feature>. The expected SHA is captured BEFORE rebase to defeat lease-weakening from background fetches (rubber-duck Q2).
• Conflict abort: git merge --abort (guarded by MERGE_HEAD) → git rebase --abort (guarded by Test-RebaseInProgress, which checks BOTH rebase-merge and rebase-apply dirs via git rev-parse --git-path).
• New pre-flight Test-WorktreeBlocked rejects with new error code worktree_dirty if any in-progress merge/rebase/cherry-pick/revert OR uncommitted changes are detected (rubber-duck Q6).
• Edge case: rebase no-op (HEAD didn't move despite behind_by>0) exits success without force-push.
• Initial-push path: if origin/ doesn't exist yet, falls back to git push --set-upstream (no lease — nothing to overwrite).
• Envelope: new strategy: 'rebase', old_head_sha, new_head_sha fields. Dropped merge_commit_sha (rebase has no single integration SHA).
• error_code value merge_conflict retained for workflow-route + lint contract compatibility. The semantic meaning is "drift integration conflicted"; only the operator-instructions in the conflict-gate prompt change.

Workflow (feature-pr.yaml):

• version: 2.4.9 → 2.4.10
• Agent comment updated: rebase rationale, force-with-lease safety argument.
• Conflict-gate operator instructions: full rebase recipe with explicit git rev-parse refs/remotes/origin/<feature> for the lease, rebase --continue loop, force-with-lease push.
• Failed-gate prompt: documents new worktree_dirty cause + updated push_failed (lease-refused) semantics.

Tests (integrate-target-drift.Tests.ps1):

• Clean-drift: now asserts strategy='rebase', new_head_sha != old_head_sha, origin/<feature> HEAD landed on new_head_sha (force-with-lease verified end-to-end).
• Conflict: asserts neither rebase-merge nor rebase-apply dir remains after abort.
• New Worktree dirty context.
• Envelope-contract updated for new field set.
• 6/6 pass.

Validation

• Invoke-Pester .\.conductor\registry\tests\integrate-target-drift.Tests.ps1 → 6 passed, 0 failed.
• Invoke-Pester .\.conductor\registry\tests\lint-feature-pr.Tests.ps1 → 28/29 passed. The remaining failure (missing-ado-remediation-route) is pre-existing on main and unrelated.

Rubber-duck adoption

• Q1 (rebase vs --rebase-merges): plain rebase — preserves feature-PR review clarity; MG merge trail already exists in ADO PR history.
• Q2 (force-with-lease syntax): explicit --force-with-lease=refs/heads/<feature>:<expected-sha> + HEAD:refs/heads/<feature> push refspec.
• Q3 (rebase-detection): Test-Path (git rev-parse --git-path rebase-merge) OR rebase-apply (different rebase modes use different dirs).
• Q4 (operator instructions): included git fetch origin <target> <feature> pre-step + explicit lease.
• Q5 (divergence gate): kept (more critical with force-push than with merge).
• Q6 (pre-checks): added in-progress merge/rebase/cherry-pick/revert + uncommitted-changes guard up front.

Closes the user-feedback gap on PR #471.