fix: normalize_url preserves path/query casing, lowercasing only scheme and host

[enriquealonso01]

Summary

normalize_url lowercased the entire URL via str(yarl_new_url).lower(). Since only the scheme and host are case-insensitive (RFC 3986 §6.2.2.1), this silently merged distinct URLs during request deduplication — e.g. https://example.com/Product/ABC and https://example.com/product/abc collapse into one request key, and case-significant query tokens (signed URLs, API tokens) were mangled.

Changes

• normalize_url now returns str(yarl_new_url) unchanged: yarl already lowercases scheme and host during parsing, while path/query/fragment keep their original casing
• Docstring updated to state the case-sensitivity contract

Closes #2008

Test plan

• Updated the existing HTTPS://EXAMPLE.COM/?KEY=VALUE case: scheme/host lowercase, query casing preserved
• Added regression cases: path casing preserved (/Product/ABC), query token casing preserved (?token=SeCrEt)
• pytest tests/unit/_utils/test_requests.py — 17 passed

Breaking change note

As flagged in #2008 (comment) (this is planned for v2 as breaking): anyone who relied on full-lowercase unique keys will see different keys for case-distinct URLs. Targeting master per the current default branch — happy to rebase onto a v2 branch if one exists.

Submitted as part of Hacktoberfest-adjacent open-source contributions by @enriquealonso01.