Fix adoption P1 and P2 evidence gaps

[liolevx]

Summary

• recover bounded implementation context across the pinned misses
• distinguish workspace and instruction authority without silent takeover
• retain actionable bounded failure diagnostics
• clarify first-task setup and record the bounded learning/Director experiment

Verification

• format, lint, typecheck, 227 tests (2 skipped), build
• built safety and packaged CLI smoke
• checkpoint safeguards on Windows and WSL
• independent review approved

Scope

No P3 expansion, publish, or productivity/universal-compatibility claim.