docs(state): qualify legacy lusoris/vmaf issue + PR refs after VMAFx migration

[lusoris]

Summary

Cross-referenced docs/state.md against VMAFx/vmafx GitHub Issues.

Audit finding: VMAFx/vmafx carries 0 issues (all 179 open items returned by the API are PRs). The historical bug-tracker numbers cited in state.md (#239, #857, plus resolving PR refs #241, #310, #870) live on the archived lusoris/vmaf repo and now collide with unrelated PR numbers on VMAFx/vmafx:

Cite in state.md	Original (lusoris/vmaf)	Current VMAFx/vmafx #N
Issue #857	cambi_cuda SIGSEGV bug report	(no such issue — #857 returns 404)
Issue #239	Vulkan wall-clock serialisation report	(issues are zero — #239 is PR "fix(python): pyx cdef extern from…")
PR #241	T7-29 v2 async pending-fence	PR "fix(test): sunset VmafLegacyQualityRunner test methods"
PR #870	fix/cambi-cuda-host-preprocessing	(no such PR — #870 returns 404)
PR #310	ADR-renumber dedup sweep	(different PR on VMAFx)

This PR qualifies every bare cite as lusoris/vmaf#NNN so future maintainers don't follow ambiguous numbers to a different PR on the active repo.

No GitHub issues were closed and no rows added — repo state is already aligned (the archived repo's issues are already closed; the new repo never imported them). This is documentation hygiene only.

Counts

• Issues closed on GitHub: 0 (none open; archived repo cannot be touched)
• Rows added to state.md: 0 (no live issues to track)
• Rows updated in state.md: 5 (qualified cites: 1 header banner line, 2 in Recently-closed table, 2 in Confirmed-not-affected duplicate rows)
• Banner line added: 1 (audit timestamp + rationale at line 115)
• New changelog fragment files: 1 (changelog.d/changed/state-md-github-issues-crossref.md)

Reproducer

# Verify the repo really has zero issues
gh api "repos/VMAFx/vmafx/issues?state=open&per_page=100" --paginate \
  | jq -r '[.[] | select(.pull_request == null)] | length'
# -> 0

gh api "repos/VMAFx/vmafx/issues?state=closed&per_page=100" --paginate \
  | jq -r '[.[] | select(.pull_request == null)] | length'
# -> 0

# Verify the historical numbers resolve to the archived repo
gh api repos/lusoris/vmaf/issues/857 --jq '"\(.title) state=\(.state)"'
# -> "cambi_cuda segfaults on every input (rebuilt 2026-05-15) state=closed"

# Verify VMAFx PR #241 is unrelated to the cited Vulkan work
gh api repos/VMAFx/vmafx/pulls/241 --jq '.title'
# -> "fix(test): sunset VmafLegacyQualityRunner test methods per ADR-0749"

# Verify no bare unqualified historical cites remain in state.md
grep -nE 'Issue #(239|857)|^\| \*\*#(239|857)\*\*|PR #(241|310|870)' docs/state.md | grep -v 'lusoris/vmaf'
# -> (empty)

Bug-status hygiene (ADR-0165)

• docs/state.md updated in this PR — five citation qualifications + one audit banner.

Netflix golden-data gate (ADR-0024)

• I did not modify any assertAlmostEqual(...) score in the Netflix golden Python tests.

Deep-dive deliverables (ADR-0108)

• Research digest — no digest needed: docs-only citation-qualification sweep with no design surface.
• Decision matrix — no alternatives: only-one-way fix (qualify ambiguous cites; leaving them ambiguous is the bug).
• AGENTS.md invariant note — no rebase-sensitive invariants: docs-only change to a fork-local file.
• Reproducer / smoke-test command — pasted above under "Reproducer".
• CHANGELOG fragment — changelog.d/changed/state-md-github-issues-crossref.md.
• Rebase note — no rebase impact: docs/state.md is fork-local and not in the upstream Netflix tree.

🤖 Generated with Claude Code