[prestoncabe]

Screeners can now include a Next steps form component. It shows application instructions and contacts for eligible benefits, unmet requirements and an analyst note for ineligible benefits, and the benefits that still need more information. Analysts can show decisions as they arrive or wait until every benefit is decided.

Closes #556. Fourth PR for #552, following #574. Print summary and AI draft integration follow in #557 and #558.

The editor loads benefit sources in the background and copies their content into each new component's schema configuration. Only a new component's content editing and form saving wait for its pending copy; other editor work stays available. A modal lets analysts edit that screener's copy or reset an individual benefit from its source. Missing copies are flagged, removed benefits are ignored, and importing a schema preserves its saved content. Publishing uses the existing form-schema snapshot; public viewers do not fetch working benefit content.

Publishing preserves benefit IDs in the published subcollection so evaluation results match the schema's copied content. Screener transfers and example imports remap those content keys when allocating new benefit IDs, preserving analyst edits.

Markdown is sanitized without template evaluation. HTTP(S) links open in a new tab; email and phone links use their contact apps. Pending or failed evaluations suppress stale instructions. Visible Next steps fields announce completion and replace the page's completion message; hidden or absent fields retain that fallback.

Modal saves preserve focus and other benefits' latest content, and discarding unsaved modal edits requires confirmation. Form saves prevent duplicate requests and keep newer edits marked unsaved. Deferred copies survive undo/redo and retries, follow the original component object, and never rewrite imported components that reuse an ID.

Publishing requires a saved form and rejects a missing artifact before changing publication metadata. Storage write failures reach the caller, save and publish errors remain visible for retry, and saving a new unedited form stores its initialized schema. Publication timestamps are recorded after the artifact work succeeds.

The rebase brings in the updated icon library from main. Individual icon imports prevent the development server from loading the library's entire icon source tree, which exhausted Chromium resources during repeated full-page navigation in CI. Four consecutive page loads and evaluations passed locally with seven distinct icon modules and no resource errors.

Validation:

• All 285 frontend tests passed across 47 files, including save/publish errors, initial form saves, save state, late-loaded modal content, pending copies, undo/redo, retries, and imported component IDs.
• Production frontend build passed.
• All 235 builder API tests passed, including missing-form rejection, storage failures, publication ordering, and the nested component import regression.
• Both public-screening Playwright tests passed locally, covering display modes, partial completion, eligibility changes, safe Markdown, pending evaluations, and failures.
• Verified the viewer layout, content modal, and properties-panel edits in Chromium.
• Authenticated end-to-end coverage passed in CI for copying, editing, resetting, saving, publishing content, and matching published benefit IDs.
• All five CI checks passed on the rebased head with the icon-loading fix: frontend, builder API, library API, end-to-end, and documentation. Browser coverage includes recovery from an unsaved form by saving and publishing it.
• git diff --check passed.