v2 re-baseline: unshallow, pin upstream v2.1.3, record the delta #124

Parent: Phase 1a item P1.0 (plugin port). Must precede every port change.

Why

The shallow clone already caused one false ancestry conclusion, and every planning
claim downstream (P1.3–P1.9) needs the pinned target and the exact fork/upstream
delta as ground truth. The evaluation verified the numbers only against
opencode-skein@dev (5fea944430) and upstream/dev (09a1db3, opencode v2.1.3);
a tagged release is the deliverable, not branch drift. This is mechanical, but it is
the lock everything else clips into.

What Changes

1. git fetch --unshallow upstream on ~/dev/opencode-skein and assert bounded
   ancestry (rev-list --count) before any claim that cites it.
2. Pin the target release: v2.1.3 (tag, not dev), recorded in
   docs/skein-2/00-current-state.md and every port change's tasks.
3. Generate the fork/upstream delta map: every fork-only source tree and every
   upstream file the fork cuts into (the evaluation's two lists, regenerated fresh
   against the pinned tag) — written into docs/skein-2/00-current-state.md.
4. Record the fork-specific mutable state (config/state dirs, data formats) for the
   migration decision in adopt-upstream-acp-and-service.

Non-goals

• No code changes. No behavior change. No architecture decision here (that is
  P1.1's ADR 13).
• No attempt to rebase the fork (the 500–700 commit gap stays; retirement, not
  rebasing, is the plan).

Risks

• Ancestry assertions made against a still-shallow repo silently pass wrong —
  hence the unshallow step is part of validation, not a prerequisite to it.
• The delta map regenerates differently across agents; make it a script output
  (git diff --stat, name-only), committed as data, never hand-copied.

Impact

Planning docs update; no runtime impact. Gates K8 (shallow clone) closed.