Fix historical URL filtering and Common Crawl stream cleanup - #11

Summary

Hotfixes two issues reproduced by the post-release 0.6.0 functional audit.

Historical Intelligence

Malformed archived URLs such as bracketed non-IP hosts can make urllib.parse.urlsplit raise ValueError. The filter layer now treats those unusable archived URLs as dropped input instead of aborting the whole historical analysis.

Common Crawl lifecycle

When a consumer stops early (for example because --max-results is reached), nested Common Crawl async iterators are now explicitly closed with contextlib.aclosing. This prevents unraisable async-generator shutdown errors such as aclose(): asynchronous generator is already running.

Regression coverage

• malformed bracketed archived host is safely filtered
• historical analysis skips malformed captures and continues
• early Common Crawl provider close cascades to the nested stream

This PR intentionally contains no new features.