fix(url): fall back to Latin-1 when a response has no declared charset

[NavidSassan]

Problem

url.fetch() decodes every response body as strict UTF-8 when the server sends no charset (response_charset is None → default 'UTF-8'). A single non-UTF-8 byte then aborts the whole check with a UnicodeDecodeError.

Real-world trigger: Comet System Web Sensors serve the degree sign ° as the raw ISO-8859-1 byte 0xb0 in their values.json and declare no charset, so cometsystem fails with:

'utf-8' codec can't decode byte 0xb0 in position 164: invalid start byte while fetching http://.../values.json

This is a regression from the pre-httpx engine (lib 3.x), which decoded tolerantly.

Fix

When the server declared no charset and the UTF-8 assumption fails, fall back to Latin-1, which maps every byte 0x00–0xFF one-to-one and can never raise. 0xb0 becomes °, which is what the firmware means.

A charset the server explicitly declared still fails on mismatch, so genuine encoding errors are not masked (the existing test_decode_error_returns_failure guards this).

Why Latin-1 and not surrogateescape

The plugins decode a body, build a message, and print it to stdout (re-encoding to UTF-8). Tested against the byte values seen across the monitoring-plugins decode issues (0xb0, 0x81, 0xc2, 0xfc):

bytes	latin-1 → output	surrogateescape → output
25.3 \xb0C	25.3 °C OK	\udcb0 UnicodeEncodeError
>m\x81ller	OK	UnicodeEncodeError
M\xfcller	Müller OK	UnicodeEncodeError

surrogateescape only relocates the crash from fetch to output. Latin-1 maps to real Unicode scalars that survive both the JSON round-trip and the stdout re-encode. This matches how the lib already fixed the Windows console decode crashes (shell.py decodes with the actual OEM code page rather than papering over a wrong-codec decode with an error handler).

Tests

• New test_undeclared_charset_falls_back_to_latin1 reproduces the cometsystem case (body b'25.3 \xb0C', no charset → '25.3 °C').
• Existing test_decode_error_returns_failure stays valid: a declared charset with invalid bytes still fails.

Full tests/url suite green (90 tests), ruff check/ruff format clean.

Fixes the cometsystem decode failure reported downstream in Linuxfabrik/monitoring-plugins.