fix(gp2rs): write arrangement XML as UTF-8 so GP import survives non-ASCII metadata (#984)

The GP→arrangement-XML writers persisted their output with
`Path.write_text(xml_str)` and no explicit encoding. On Windows that uses
the cp1252 default, so a non-ASCII metadata character — e.g. the © in an
album name like "Chrysalis©1982" — was written as the lone byte 0xA9.
The XML is read back as UTF-8 (expat's default), where 0xA9 is an invalid
start byte, so `parse_arrangement` died with:

    xml.etree.ElementTree.ParseError: not well-formed (invalid token): line 10, column 22

and the whole Guitar Pro import failed (HTTP 500). All three arrangement
XML writes (gp2rs.py, gp2rs_gpx.py ×2) now pin encoding="utf-8".

CI runs on Linux (UTF-8 default) so the bug was invisible there; the new
test pins the locale-independent contract at the source level plus a
round-trip of a © album name.


Claude-Session: https://claude.ai/code/session_01EBQCHCNA81E9tHmSDHSe2Q

Signed-off-by: ChrisBeWithYou <christian.a.cowan@gmail.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

tests/test_gp2rs_xml_encoding.py

@@ -0,0 +1,53 @@
+"""Regression: the GP→arrangement-XML writers must pin UTF-8.
+
+A bare ``Path.write_text(xml_str)`` uses the platform's *default* text
+encoding. On Windows that is cp1252, which encodes a non-ASCII metadata
+character — e.g. the © in an album name like "Chrysalis©1982" — as the lone
+byte 0xA9. The XML is then read back as UTF-8 (expat's default), where 0xA9
+is an invalid start byte, so parsing dies with
+
+    not well-formed (invalid token): line N, column 22
+
+CI runs on Linux (UTF-8 default), so the bug is invisible there and a plain
+functional test would pass on the old code too. These assertions instead pin
+the locale-independent contract directly.
+"""
+
+import inspect
+import re
+import xml.etree.ElementTree as ET
+
+import gp2rs
+import gp2rs_gpx
+
+
+def test_arrangement_xml_writes_specify_utf8():
+    # Every write of the arrangement XML string must pass encoding="utf-8"
+    # so non-ASCII metadata survives regardless of the host locale.
+    for mod in (gp2rs, gp2rs_gpx):
+        src = inspect.getsource(mod)
+        bare = re.findall(r"\.write_text\(\s*xml_str\s*\)", src)
+        assert not bare, (
+            f"{mod.__name__}: XML write must pass encoding=\"utf-8\" — a bare "
+            f"write_text() uses the platform default (cp1252 on Windows) and "
+            f"mangles non-ASCII metadata into invalid UTF-8"
+        )
+        assert 'write_text(xml_str, encoding="utf-8")' in src, (
+            f"{mod.__name__}: expected a UTF-8-pinned arrangement XML write"
+        )
+
+
+def test_utf8_write_round_trips_non_ascii_album():
+    # The behavioural end of the contract: a © album name written as UTF-8
+    # parses cleanly and reads back intact (the cp1252 write does not).
+    from pathlib import Path
+    import tempfile
+
+    xml_str = (
+        '<?xml version="1.0"?>\n<song>\n'
+        "  <albumName>Chrysalis©1982</albumName>\n</song>\n"
+    )
+    path = Path(tempfile.mkdtemp()) / "arr.xml"
+    path.write_text(xml_str, encoding="utf-8")
+    root = ET.parse(path).getroot()
+    assert root.findtext("albumName") == "Chrysalis©1982"