[tannevaled]

Summary

M6 of the go-wasmcontainers WebGPU-readiness roadmap (M1 shipped as #341). The peer's own ask: "Fetch + RenderDocument + JSLog + the DOM... provided a render can wait for the page to post 'done', meaning async completion with a timeout." No WebDriver/CDP protocol needed, and no new goroutine — this rides entirely on the existing settle loop and JSTimeout/deadman budget.

• New Engine.EnableDoneSignal bool (default false, no behaviour change for any existing caller — not even an extra global name exists).
• When set, js/window.go's install() installs a global __webengineDone(value) a page's own script can call once its async work is finished (first call wins; later calls are no-ops).
• settle_scripts.go's settle() pass loop now breaks as soon as sess.Done() reports the signal fired, instead of running out maxSettlePasses/JSTimeout.
• New Session.Done() (value interface{}, ok bool) (js/session.go) reports the signal and the page's own value, exported via goja the same way Eval's return value is.

Closes #346.

Test plan

• go build ./..., go vet ./... clean
• enginejs_test.go's new TestEnableDoneSignalStopsSettleEarly reuses TestSettleFixpointCap's own self-replicating script chain: with the option on, the loop stops right after __webengineDone() fires (data-d="2"); with it left off, the identical page still reaches data-d="4" — proving the feature is a true no-op for existing callers.
• js/js_test.go's new TestSessionDoneSignal covers Session.Done() directly: a call with a value, no call at all, a call with no argument (goja's Undefined(), not Go nil), and — with the option off — the same call throwing ReferenceError.
• Full suite green, no coverage-gate regression (engine 95.8%, js 94.6%→94.9%).

🤖 Generated with Claude Code