perf(cookies): parallelize browser strategy queries

[mherod]

Summary

• Replace sequential for...of await loop with Promise.allSettled in batchQueryCookies so Chrome, Firefox, and Safari are queried in parallel
• Each strategy targets independent browser databases with zero data dependency
• Error handling behavior unchanged (continueOnError flag respected)

Closes #485

Test plan

• Type-check passes
• Biome lint passes (225 files, 0 errors)
• All 588 tests pass
• Pre-commit and pre-push hooks pass

[claude]

The parallelization change is correct and well-structured.

What the PR does well:

👍 Promise.allSettled is the right primitive — it guarantees all three browser strategies are always attempted, even if one fails early. Using Promise.all here would have short-circuited on the first failure when continueOnError = false.

👍 Error handling semantics are preserved — the continueOnError flag is threaded through both the inner queryStrategy calls and the new outer settlement loop.

👍 Independent I/O is the correct justification — Chrome, Firefox, and Safari each read from completely separate database files, so parallel access is safe and there are no shared resources or ordering constraints to worry about.

One minor observation worth noting (non-blocking):

🔵 "rejected" branch in the outer loop is a narrow path when continueOnError = false — queryStrategy already catches internally and re-throws when continueOnError = false, so Promise.allSettled captures that throw as a rejected result. The outer else { throw result.reason } then re-throws it. This is functionally correct, but it means an uncaught exception in queryStrategy (e.g., from something outside the try/catch block like the strategy constructor throwing) would also land here. The current code handles that correctly, so this is purely an observation rather than a bug.

🔵 No direct unit tests for the parallel behavior in batchQueryCookies itself — existing tests mock out batchQueryCookies entirely. A test verifying that all three strategies are invoked concurrently (e.g., using spies to detect overlapping execution) would add confidence, but the existing integration path through batchGetCookies covers the happy path.

Overall the change is minimal, well-reasoned, and the test plan confirms all 588 tests pass. Ready to merge.