v1.108.141 — gold corpus v2: TypeScript and Go join the measurement

Added

• The channel-accuracy gold corpus now spans three languages (authored-scenarios-v2). TypeScript scenarios mirror the Python design with TS-native semantics: extends-declared notifier subclasses plus a UI-toast Notifier homonym trap (AST); a duck-typed PushGateway.send conformer against ConveyorBelt.send and ElevatorBank.send physical-dispatch traps; @on("invoice_paid") handler methods against @get route-path and @task job-name substring traps. Go scenarios exercise the duck channel exclusively — deliberately, since Go's implicit interfaces have no declared inheritance for the AST channel to walk and Go has no decorators: RingBuffer.Push / PriorityList.Push conformers against Lawnmower.Push (physically pushing the mower) and Recruiter.Push (advocating a candidate) traps.
• Measured round 2 (aggregate, 3 languages): AST precision 0.818, duck 0.5, decorator 0.556 — recall 1.0 across every channel and language; the artifact gains a per_language breakdown. The duck prior (0.65) now reads clearly optimistic against the harder cross-language corpus (0.5 measured) — exactly the drift the measured_ref exists to make visible. Operating constants stay declared pending a deliberate replay-gated recalibration.
• Harness: corpus hash covers .py/.ts/.go; per-language stats; the unlabeled-pair guard and the CI re-measurement gate carry over unchanged.