Add onBrowserBailout Fizz option

[gnoff]

Adds a new Fizz option, onBrowserBailout, for observing intentional server-render bailouts caused by ReactDOM.browser() and future APIs that use the same recoverable error mechanism. The callback receives the original recoverable error and ErrorInfo, defaults to a noop, and runs only when Fizz successfully recovers by deferring work to the browser.

Recoverables consumed within Suspense or used to abort recoverable boundaries are reported through onBrowserBailout without also invoking onError. A bailout outside Suspense remains fatal and reports only through onError, with the original recoverable preserved as its cause, while directly throwing the value continues to behave like a normal render error.

Plumbs the option through the streaming, resume, and prerender entry points for Node, browser, Edge, Bun, FB, markup, and noop renderers while preserving the positional Fizz request API for callers that do not expose the option.

Uses an environment-neutral browser-only rendering message for the isomorphic browser() value and updates the production error mapping. Tests cover successful browser bailouts, recoverable abort reasons, root-fatal behavior, component stack information, the default noop behavior, and direct throws in development and production.

[eps1lon]

and runs only when Fizz successfully recovers by deferring work to the browser.

Are we testing this at the moment? Sounds like onBrowserBailout shouldn't run if a sibling in the same Suspense boundary throws.

[eps1lon]

Are we calling onBrowserBailout for each unique browser() invocation that causes a boundary to be deferred or each use callsite?

[gnoff]

Are we testing this at the moment? Sounds like onBrowserBailout shouldn't run if a sibling in the same Suspense boundary throws.

When you error a boundary in Fizz it's first one wins, we don't really consider a second error to be meaningful so if use(browser()) is the first error within a boundary it'll get reported and you won't observe any pending work that errors too. We could test but I don't know if it is meaningful since in Fizz browser() is really just an error with slightly different reporting mechanics and we could potentially change this in the future. (i.e. I think it can be unspecified behavior vs a public contract)

Are we calling onBrowserBailout for each unique browser() invocation that causes a boundary to be deferred or each use callsite?

We call it once per Boundary and it's first use wins (of course other errors win "more" if they are first per above)

[react-sizebot]

The size diff is too large to display in a single comment. The GitHub action for this pull request contains an artifact called 'sizebot-message.md' with the full message.

Generated by 🚫 dangerJS against dde3dad