fix: preserve path and query casing in normalize_url (#2008)

[aryansk]

Summary

Fixes #2008 — normalize_url lowercased the entire URL, including the path and query, so any two URLs differing only in path or query casing produced the same unique key:

• https://example.com/Product/ABC and https://example.com/product/abc → same key
• https://example.com/?token=SeCrEt and https://example.com/?token=secret → same key

Per RFC 3986 §6.2.2.1 only the scheme and host are case-insensitive — the path and query are case-sensitive. On sites with case-sensitive paths (base64/hashid identifiers, usernames), case-distinct pages were silently deduplicated with no log message or statistic revealing it.

Change

yarl already lowercases the scheme and host on parse (verified empirically: URL('HTTP://Example.COM/Product/ABC?Token=SeCrEt') → http://example.com/Product/ABC?Token=SeCrEt). The fix removes the final .lower() on the constructed URL so path/query/fragment casing survives normalization, matching browser behavior. The docstring now states precisely which parts are case-folded.

Breaking change note: this changes how default unique keys are computed — crawls that relied on case-insensitive dedup will now visit more pages, and keys stored in persisted queues won't match newly computed ones. Per the issue, this should land in a major (2.0) release with the change documented loudly.

Validation

• New parametrized cases: scheme/host still lowercased (HTTPS://EXAMPLE.COM/?KEY=VALUE → https://example.com/?KEY=VALUE), path/query/fragment case preserved, trailing-slash removal unaffected.
• New regression test: case-distinct paths normalize to different URLs (dedup no longer collapses them).
• tests/unit/_utils — 299 passed, ruff check + format clean.