Fix parse() raising TypeError on text-mode file-like objects

[afonsojanu]

Related to #427.

_open_resource() returns whatever a file-like object's .read() gives back untouched. For io.BytesIO that's bytes, which is fine, but for io.StringIO it's a plain str, and convert_to_utf8() always matches its input against a bytes regex (RE_XML_PI_ENCODING), so it blows up with:

TypeError: cannot use a bytes pattern on a string-like object

The parse() docstring actually recommends wrapping an untrusted string in io.StringIO or io.BytesIO, so this hits anyone following that advice with the text variant. Passing a plain str directly already works, since that path already encodes to utf-8 a few lines further down — this just applies the same treatment when the data comes from a text-mode file-like object.

Fix is a small change in _open_resource(): if the object we read from a file-like isn't bytes, encode it, matching the existing plain-string branch. Added two tests: one directly on _open_resource(), one going through parse() with an io.StringIO. Confirmed both fail with the original TypeError without this change and pass with it; full existing suite (4335 tests) still passes.

Targeting releases rather than main since this is the 6.0.x maintenance branch where the bug actually lives — I noticed main already handles this correctly after the _to_in_memory_file() rework, so this is just a small backport for people on the current stable line.