fix: default read_html() encoding to UTF-8

[Copilot]

read_html(url) double-encoded UTF-8 characters on Windows (codepage 65001) because htmlReadMemory received NULL as the encoding hint, causing libxml2 to fall back to the system locale — producing garbage like "Ã\u0084pfel" instead of "Äpfel". Regression vs. xml2 1.3.6, introduced via the Rtools libxml2 upgrade.

Changes

• R/xml_parse.R: Change encoding default from "" to "UTF-8" in read_html(), read_html.default(), and read_html.response(). Documents that declare <meta charset> are unaffected — libxml2's HTML parser still re-parses with the declared encoding.
• tests/testthat/test-xml_parse.R: Add regression test — round-trips "\u00c4pfel" through charToRaw() → read_html() → xml_text() without explicit encoding.
• NEWS.md: Changelog entry.

# Before (broken on Windows codepage 65001):
xml_text(xml_find_first(read_html(url), "//div"))
#> [1] "Ã\u0084pfel"   # double-encoded

# After:
xml_text(xml_find_first(read_html(url), "//div"))
#> [1] "Äpfel"          # correct