fix: include live Chrome history in SQLite snapshots

[Iams4kura]

What does this PR do?

Chrome can commit new history rows to its SQLite write-ahead log before checkpointing them into the main History file. ChromeHistoryReader copied only that main file, so indexing a live profile could silently omit recent pages.

This replaces the main-file copy with SQLite's online backup API. The reader now snapshots a consistent view of both the main database and committed WAL content into memory, keeps the browser database read-only, and closes both connections on every path.

Regression evidence

• Before: uv run --no-project --offline --with pytest pytest tests/test_chrome_history_reader.py -q exited 1; the reader returned zero documents while the committed recent row remained in the WAL.
• After: the same command exited 0 and returned the recent page.

Verification

• uv run --no-project --offline --with pytest pytest tests/test_chrome_history_reader.py -q
• uvx --from 'ruff==0.12.7' ruff check packages/leann-core/src/leann/readers.py tests/test_chrome_history_reader.py
• uvx --from 'ruff==0.12.7' ruff format --check packages/leann-core/src/leann/readers.py tests/test_chrome_history_reader.py
• python3 -m py_compile packages/leann-core/src/leann/readers.py tests/test_chrome_history_reader.py
• git diff --cached --check

The full dependency environment could not be synchronized locally because repeated PyPI downloads timed out; the deterministic reader regression and all changed-file checks above completed successfully.

Related Issues

No existing issue or pull request matched this failure after searches for ChromeHistoryReader, the legacy temporary filename, browser-history WAL handling, and SQLite backup behavior.

Checklist

• Focused regression test passes
• Code formatted (ruff format --check)
• Changed files pass lint (ruff check)
• Full pre-commit suite (pre-commit run --all-files) — dependency downloads timed out; equivalent changed-file Ruff and diff checks pass