[huangruiteng]

Summary

Fixes the post-merge P2 from #6987: when a Goal has goal_objective plus cumulative outcomes or configured checks, the shared goal-acceptance inspect command must carry adoption guidance and the existing acceptance obligations. The prior command-level deduplication kept only the first reason.

This change keeps one read command, merges all applicable reason clauses, preserves an existing custom reason, and remains idempotent. It does not add another inspect call or a second acceptance authority.

Verification

• Full control-plane TS suite: 5236 tests, 5163 passed, 73 PostgreSQL-condition skips, 0 failed.
• Focused interaction-contract and TurnEnvelope suite: 59/59 passed.
• Real File and SQLite CLI matrix after objective adoption: 4/4 passed across configured checks and cumulative outcomes; inline work_context.sources retained both guidance sets with no extra required inspect.
• Scheduler boundary expectations: 3/3 passed after updating the stale pre-existing goal-state guidance expectation.
• Guidance/output budget checks: 10/10 passed.
• npm run -s typecheck:control-plane, Ruff, Python compile and git diff --check passed.
• loopx canary premerge --from-git-diff: passed, 18/18 selected checks, 0 failures, 0 manual holds.
• Read-only local loopx-meta status and acceptance inspection succeeded. The active meta Goal currently has no objective/outcomes/checks, so the exact branch was exercised in isolated real File/SQLite Goals instead; no active Goal state was modified.

Residual risk

No live-model adoption or benchmark-effect experiment was run. PostgreSQL tests remained conditionally skipped in the TS suite; this diff does not change provider storage behavior and the real provider regression covers File and SQLite.

Related: #6987 post-merge audit discussion discussion_r4335733624.