When a C file references another source file via `/* in file.c */`, the
parser read it with bare `File.read` which uses
`Encoding.default_external`. On systems where this is US-ASCII (e.g.
Debian CI), non-ASCII bytes in the source file cause `ArgumentError:
invalid byte sequence in US-ASCII` in String#scan.

Use `RDoc::Encoding.read_file` instead, which reads in binary mode and
properly handles encoding detection and transcoding.

This was triggered by Ruby commit
[`a2531ba293`](ruby/ruby@a2531ba293) which
added UTF-8 right arrows (→) in comments in `class.c`, which is
referenced from `object.c` via `/* in class.c */`.