fix: screenreader dropdown announcements and option navigation

[k-gerner]

Why:

• VoiceOver could produce interrupted or duplicate speech when dropdown state, result counts, and active options changed together.
• Highlighted matches were split across visual spans, causing assistive technology to pronounce option values as separate fragments.
• SearchBar and FilterSearch require different navigation behavior because FilterSearch automatically selects its first option.

Changes:

• Replace remounted announcement content with a stable polite status region that clears pending text and cancels stale delayed updates.
• Separate one-time usage instructions from result-count announcements and suppress result counts while users navigate options.
• Report the combobox as expanded only when the dropdown is active and contains a listbox.
• Keep DOM focus on the combobox and identify the highlighted option through aria-activedescendant.
• Preserve typed SearchBar text while navigating suggestions and commit the highlighted value only when selected.
• Preserve FilterSearch’s automatic first selection while making Arrow Down advance to the following option.
• Render dropdown entries as listbox options with selection state and complete accessible names.
• Hide visually split highlighted markup from assistive technology so each option is announced as one phrase.
• Remove unnamed grouping semantics from SearchBar previews and unlabeled FilterSearch sections.
• Avoid announcing FilterSearch result counts until an actual filter-search response exists.
• Correct the plural result-preview translation from “recent previews” to “result previews.”
• Make Enter consistently submit and select the active option’s value.

Tests:

• Add ScreenReader coverage for stable nodes, delayed updates, and stale-announcement cancellation.
• Expand Dropdown tests for combobox state, active descendants, option semantics, selection state, and keyboard behavior.
• Expand SearchBar and FilterSearch tests for complete accessible names, query preservation, automatic selection, result counts, and one-time instructions.

Known gaps:

• On the Universal SearchBar page, VoiceOver queues the usage instructions but evicts them when the expanded-listbox and result-count announcements arrive.
• The current implementation marks instructions as announced even when VoiceOver does not finish speaking them, so they are not retried.
• The fixed result-announcement delay reduces contention but cannot guarantee speech ordering across browsers, screen readers, or speech rates.
• Automated DOM and axe tests verify accessibility semantics but cannot confirm the final spoken sequence; manual VoiceOver verification remains required.
• A follow-up could combine the one-time instructions with the first result count in one atomic announcement.

J=WAT-5482
Mini Demo:

search-ui-react-screenreader-update.mov

[github-actions]

Current unit coverage is 89.46337641989815%
Current visual coverage is 69.28934010152284%
Current combined coverage is 91.88461538461539%

[coveralls]

coverage: 84.912% (+0.2%) from 84.727% — screen-reader-rework-5482 into main