Adopt upstream ACP + opencode service, retire the fork (P1.7) #130

Parent: Phase 1a item P1.7. Slots after the ports (P1.3–P1.6). This change is where
the fork boundary is crossed; the verification contract (charter §6.4, §7.6) governs
it.

Why

The v2 pivot exists because carrying a 241-file fork costs a restructure-port per
upstream release. After P1.3–P1.6 the remaining fork code is acp/ (superseded — we
adopt upstream's), context/ and fork/ residue. Cutting over is justified only if
upstream demonstrably covers the two verification contracts:

• §6.4 (ACP): the external-client test list. The unit-level migration on
  skein2/acp-sdk-1.8 proved the API shape; upstream adoption must prove the
  behavior via acp-external-verification (Scenario A).
• §7.6 (lifecycle): the ten assertions (start without TUI, survive TUI close,
  attach second client, dual observation, multi-worktree sessions in one project scope,
  durable state across kill, restart reconciles process+repo truth, no re-owned task,
  abandoned worktree preserved, acceptable long-run resource profile). The evaluation
  said explicitly this was not check'd; it must be driven live before residues are
  deleted.

What Changes

1. Drive the §7.6 ten assertion against upstream opencode service live, and the §6.4
   list against upstream's ACP via acp-external-verification. Failures become either
   reasoned upstream PRs or documented gaps with mitigations — never a new fork.
2. Retire in-fork remnants per the decision (charter §19-revised): delete acp/,
   context/, fork/ and the ported areas (local/, peer/, loop/-.Tree,
   beads/) as recorded by the port changes' dispositions; the retained-delta list
   (deliberate deviations, if any) shrinks to the recorded minimum, each with a
   documented upstream route or justification.
3. Archive the skein2/acp-sdk-1.8 branch and its change artifacts; the often-cited
   docs (01-acp-migration.md) remain as historical record, marked as sunk cost.
   (Evaluation decision 3.)
4. Swap the operator's running opencode over: the fork binary installs
   (1.18.18-dev in ~/.local/bin/opencode) are replaced after the new-plugin stack is
   proven; the migration-or-fresh decision for fork-era state dirs (from
   v2-rebaseline's R-P1 inventory) is recorded before the swap.
5. Update ecosystem docs (AGENTS.md, ~/dev/skein/docs/ECOSYSTEM.md sync-policy
   section) to the new reality: plugin repo + retained delta, no --agent flag unless
   it survives P1.7 review as an upstream-able patch.

Non-goals

• No reimplementation of anything §7.6/§6.4 confirms upstream covers.
• No upstream contribution obligations invented here beyond reasoned PRs for
  demonstrable gaps.

Risks

• Deleting fork code before verification would strand autonomy behaviors; the order
  above is enforce-only-after-evidence.
• --agent flag (src/cli/cmd/run.ts) is skein-ecosystem-critical (skein's supervisor
  calls it); if upstream lacks it, it becomes a preserved-delta item with a PR attempt.
  Removing it is unacceptable per AGENTS.md.

Impact

Fork retirement, a documented gap list, installer swap, dated residue inventory. The
platform's execution plane is upstream opencode + skein-plugin from here on.