Preserve dream lineage across manifest parent fallbacks

[Copilot]

Dream reconciliation crashed on null parent_branch values and silently indexed children as roots when lineage used alternate metadata. The repair path inherited the same gaps.

• Parent resolution

  • Resolve lineage in order:

    parent_branch → base_branch → parent_dream_id/builds_on → report.md → main
    

  • Resolve dream IDs to unambiguous remote branch refs.
  • Treat explicit null values as absent.

• Index repair

  • Apply the same fallback chain to existing mis-rooted rows.
  • Rewrite only when the parent resolves to an indexed dream.
  • Preserve existing non-root parent relationships.

• Regression coverage

  • Cover null metadata, alternate lineage keys, report fallback, true roots, and unconfirmed parents.

• Fixes update_index records a genuine child dream as a root, and crashes on parent_branch: null #27

[Xingdi (Eric) Yuan (xingdi-eric-yuan)]

Copilot
The Windows CI failure here isn't caused by this PR — it's inherited. The
windows-latest matrix was added in aee5f04 (Aug 6), but main hasn't run
tests.yml since Jun 25, so the job has never been green on main.

That said, the latest commit (63451f3) takes this PR into someone else's
lane. #26 "[Windows port] Make windows-latest CI green" is already open,
non-draft, mergeable, and green as of 364fe94.

Both PRs add a file named tests/_shell.py (#26: 128 lines, this one: 95)
and both modify 7 of the same shell-test files, so whichever lands second
conflicts.

This PR's port is also incomplete. It moved Windows from 209 failures to 24,
but the 24 that remain are exactly the files #26 patches and this one doesn't:

failures	file	covered by #26
15	test_dream_gc_sh.py	yes (+44/-5 vs +6/-2 here)
3	test_dream_setup_sh.py	yes — #26 also patches dream-setup.sh itself
2	test_pre_tool_sh.py	yes
2	test_worktree_safety.py	yes — not touched here
1	test_dream_reconcile.py	yes (worktree-gc unsafe-base, not lineage)
1	test_shadow_init.py	yes — not touched here

Suggestion: drop the Windows changes from this PR and keep it to the lineage
fix, then land #26 and rebase. The lineage change itself looks good — Ubuntu is
green, all 8 new lineage tests pass on Windows too, and I verified
_resolve_parent_branch against a corpus of real manifests: it resolves the
base_branch / parent_dream_id / builds_on cases to the correct parent
branch, keeps a bare manifest at main, and no longer raises on
parent_branch: null.

[Copilot]

Copilot
The Windows CI failure here isn't caused by this PR — it's inherited. The
windows-latest matrix was added in aee5f04 (Aug 6), but main hasn't run
...

Dropped the Windows CI changes in af4d5d6. The PR now differs from main only in the four lineage implementation/test files. The targeted suites pass (224 tests), as does the full Linux suite (1,075 passed, 8 skipped).

[Xingdi (Eric) Yuan (xingdi-eric-yuan)]

LGTM