Wrap urllib3 LocationParseError from create_connection in InvalidURL (#5744)

[jbbqqf]

Summary

Wrap urllib3.exceptions.LocationParseError raised from inside
urllib3.util.connection.create_connection in requests.exceptions.InvalidURL,
so callers see a requests-native exception instead of a raw urllib3 one. Fixes #5744.

PreparedRequest.prepare_url already catches LocationParseError at parse time
and converts it to InvalidURL. The same exception class can also be raised
later from urllib3's create_connection when the hostname survives parse_url
but contains a label that fails host.encode("idna") (e.g. a DNS label longer
than 63 characters per RFC 1035). The (_SSLError, _HTTPError) except clause
in HTTPAdapter.send does catch it — LocationParseError is a
urllib3.exceptions.HTTPError — but falls through to the trailing else: raise
and re-raises it bare. The fix adds an explicit isinstance(e, LocationValueError)
branch that wraps it in InvalidURL, mirroring the behaviour of
get_connection_with_tls_context earlier in the same method.

Reproduce BEFORE/AFTER yourself (copy-paste)

git clone https://github.com/psf/requests.git /tmp/requests-5744
cd /tmp/requests-5744
python -m venv .venv && . .venv/bin/activate
pip install -e .

# BEFORE — on origin/main, the urllib3 exception leaks.
git checkout main
python -c '
import requests, urllib3.exceptions
try:
    requests.get("http://" + "a"*64 + ".example.com", timeout=2)
except requests.exceptions.InvalidURL as e:
    print("OK requests-native:", type(e).__name__, e)
except urllib3.exceptions.LocationParseError as e:
    print("LEAK urllib3 exception:", type(e).__name__, e)
'
# Expected: "LEAK urllib3 exception: LocationParseError ..."

# AFTER — checkout this PR's branch and re-run the exact same script.
git fetch https://github.com/jbbqqf/requests fix/5744-locationparseerror-invalid-host
git checkout FETCH_HEAD
python -c '
import requests, urllib3.exceptions
try:
    requests.get("http://" + "a"*64 + ".example.com", timeout=2)
except requests.exceptions.InvalidURL as e:
    print("OK requests-native:", type(e).__name__, e)
except urllib3.exceptions.LocationParseError as e:
    print("LEAK urllib3 exception:", type(e).__name__, e)
'
# Expected: "OK requests-native: InvalidURL Failed to parse: ..., label empty or too long"

What I ran locally

$ python -m pytest tests/ 2>&1 | tail -3
========== 1 failed, 617 passed, 15 skipped, 1 xfailed, 18 warnings in 80.79s ==========

The single failure is test_proxy_error, which is environment-dependent
(it relies on a network resolution behaviour) and reproduces unchanged on
origin/main with the same setup — unrelated to this PR.

The new parametrize row in test_errors:

$ python -m pytest "tests/test_requests.py" -k "test_errors and example" -v
tests/test_requests.py::TestRequests::test_errors[http://aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa.example.com-InvalidURL] PASSED

And the same test on origin/main (regression):

tests/test_requests.py::TestRequests::test_errors[...-InvalidURL] FAILED
> urllib3.exceptions.LocationParseError: Failed to parse: 'aaaa...aaaa.example.com', label empty or too long

Edge cases

Input	Before	After	Reason
http://aaa...64a.example.com (label > 63 chars)	urllib3.exceptions.LocationParseError leaks	requests.exceptions.InvalidURL	host.encode("idna") in create_connection raises LocationParseError
http://*.example.com	InvalidURL (caught in prepare_url)	InvalidURL (unchanged)	already-existing path, validated by test_preparing_bad_url
http://fe80::5054:ff:fe5a:fc0 (zone-id-less IPv6)	InvalidURL	InvalidURL (unchanged)	already covered by test_errors
Normal URL → ConnectTimeoutError	ConnectTimeout	ConnectTimeout (unchanged)	MaxRetryError branch still catches
Normal URL → ProtocolError	ConnectionError	ConnectionError (unchanged)	dedicated (ProtocolError, OSError) branch still catches
Normal URL → urllib3 _InvalidHeader	InvalidHeader	InvalidHeader (unchanged)	branch ordered before the new one

---

PR drafted with assistance from Claude Code (Anthropic). The change was reviewed manually against requests' source. The reproducer block above is the one I used during development; reviewers can paste it verbatim.