refactor: build option-relation rulebase via copy-rulebase

[takeokunn]

Summary

Comprehensively audited every branch and worktree in this checkout
(main, feat/cl-prolog-cl-weave, feat/cl-prolog-weave,
feat/ci-ubuntu-only, feat/cl-prolog-cl-weave-integration, plus the four
local worktrees) for fixes that never made it into main, now that PR #2
landed the cl-prolog/cl-weave dogfood integration. This is the one thing
that survived the filter as a genuinely useful, verified, low-risk,
not-already-covered fix.

• make-option-relation-rulebase now uses cl-prolog:copy-rulebase
  instead of hand-reconstructing a rulebase via make-rulebase +
  rulebase-visible-clauses. copy-rulebase is the library's own
  detached-copy primitive (clauses and full runtime state — operator
  tables, flags, module/source registries), so this is both simpler and
  more robust than re-deriving the same result from a narrower accessor.
  Behavior-preserving; confirmed via the full test suite.

What else I found and why it's excluded

• feat/ci-ubuntu-only, and the a8b7cf2/0700837 commits unique to
  feat/cl-prolog-weave — no unique content; fully subsumed by what's
  already on main (or an empty checkpoint commit).
• 7b65686 "build: unify cl-weave input" (dedup a nested flake input) —
  fixes a problem specific to the old flake.nix shape (cl-prolog as a
  proper flake input with .packages.${system}.default). Main's current
  flake.nix fetches cl-weave/cl-prolog as plain (flake = false)
  sources via .outPath instead, so the duplicate-input problem this
  commit fixes doesn't exist here.
• 5bdd5e2 "fix: stabilize option relation ordering" (topological sort
  via Kahn's algorithm) — targets the same non-deterministic-comparator
  bug that PR feat: dogfood cl-prolog/cl-weave with option groups, perf, and hardening #2 already fixed on main, via a different (cheaper,
  closure-size-based) comparator. Verified main's current
  validate-option-relationships already handles a 3-level transitive
  :requires chain correctly, so re-doing the fix via topological sort
  would just be a redundant, more expensive alternative — not a gap.
• 159eb01 "ci: enforce dogfood coverage thresholds" (the other half of
  this commit, unlike the copy-rulebase change) — adds an
  sb-cover-based coverage-threshold gate to flake.nix. It calls
  cl-weave:make-coverage-summary, cl-weave:ensure-coverage-thresholds,
  and cl-weave:coverage-summary-plist, none of which exist in the
  cl-weave version main now depends on (its coverage API is now
  coverage-statistics / reset-coverage / save-coverage instead).
  Porting this would mean writing new code against an API I can't fully
  validate locally (no Linux nix builder on this machine), so it's left
  out rather than shipped unverified. Worth a follow-up if coverage
  gating in CI is still wanted.
• dependabot/github_actions/actions/checkout-7 (open PR chore(deps): bump actions/checkout from 4 to 7 #1) — routine
  Dependabot maintenance; left for Dependabot to manage rather than
  folded in here.
• The cl-cli-worktree scratch worktree — untracked files only, no
  commits, and older than every other snapshot examined (missing
  src/option-relations.lisp entirely). Nothing to recover from it.

Test plan

• sbcl --non-interactive --load tests/run-tests.lisp --eval '(cl-cli/tests:run-tests)' --quit (with CL_WEAVE_SOURCE_DIR /
  CL_PROLOG_SOURCE_DIR pointed at sibling checkouts) →
  172 passed, 0 failed, 0 errored.