fix(browser): serialize concurrent download imports

[kvnloo]

Problem

BrowserService.imports() is the only mutating per-session browser operation that does not use the existing serial(id) queue.

The download dedupe path is check-then-act:

1. look for browser-downloads/<downloadId>;
2. if missing, fetch bytes and import a file;
3. persist the dedupe row.

Two concurrent import requests can both pass step 1 and import the same download twice, leaving duplicate file rows/orphaned bytes.

Change

Route imports() through the existing per-session serial queue and move the current implementation into importsOwned().

The second caller then observes the first caller's dedupe row and returns the same signed file.

Regression

A focused test races two imports for one PDF download against a stub worker and asserts:

• both callers receive one file;
• both receive the same file ID;
• exactly one file row exists.

No download format, signing, or browser navigation behavior changes.