Show annotation counts and next/previous navigation in the file browser

### Feature summary

Provide a quick overview of file annotations and easily navigate to them.

### What problem are you trying to solve?

When a session produces inline annotations (review findings, agent comments) across a diff, there's no way to tell at a glance how many exist or where they are. Today you have to open each file and scroll to find them. On a change touching a dozen files with a handful of annotations, this means:

- **No visible count** — you can't tell whether a file has 0 or 8 annotations without opening it.
- **No total** — you don't know how much review feedback is outstanding overall.
- **No jump** — there's no "go to next annotation" affordance, so triaging means manual scrolling per file.

The result is that annotations get missed, and reviewing them is slower than it should be.

### Proposed solution

1. **Per-file badge in the file browser** — each file row shows a count of its annotations, styled by the highest severity present (blocking / warning / info). Files with zero annotations show no badge.
2. **Aggregate count in the header** — a total like `12 annotations across 5 files`, with unresolved vs. resolved broken out if resolution state exists.
3. **Next/previous navigation** — prev/next controls (plus keyboard shortcuts, e.g. `n` / `Shift+N`) that walk annotations in file-then-line order, crossing file boundaries automatically: opens the target file, scrolls the annotation into view, and focuses it.
4. **Filtering** — optionally collapse the browser to only files that have annotations, and/or filter by severity.

### Workflow impact

_No response_

### Installation context

_No response_

### Additional context

_No response_

Es werden keine Dateien, Tests oder Einstiegspunkte genannt. Beginne damit, den Dateibrowser und den bestehenden Datenfluss für Annotationen zu finden, und ermittle anschließend, wie Anzahl, Schweregrad, Auflösung, Navigation und Filterung zusammenspielen. Als abgeschlossen gilt die Aufgabe, wenn Anzahlen pro Datei und aggregierte Anzahlen sichtbar sind und Benutzer mit dem angegebenen Navigationsverhalten durch Annotationen über mehrere Dateien hinweg navigieren können.

Vom Indexierungsmodell aus dem Issue-Text verfasst.