Reading a non-text file gives a clearer, file-specific error

When Claude Code reads a file, it now checks the file's encoding (how its characters are stored) and marks the file if it is not text, such as an image or program file. Reading a marked file gives an error whose message is built for that file. Before, the error came only when a special decoding option was on and the text contained unreadable characters, and the message was always the same.

When reading only part of a file, Claude Code now returns empty content if the first chunk is not UTF-8, the standard text encoding.