Confusing behavior of selectors with buffered file objects

Using selectors with buffered file objects is confusing because the selector will only report whether there is something to read in the underlying file descriptor and ignore the internal buffer. This is especially relevant because buffered file objects are used in many places in the standard library, e.g. sys.stdin or socket.makefile().

I propose to add a corresponding warning to the docs, e.g. like this:

Selectors only look at the underlying file descriptors, so they do not know about data that is buffered in python code. Buffered file objects are used in many places in the standard library, e.g. sys.stdin or socket.makefile(). In those cases, it is recommended to bypass the buffering by using os.read(fileobj.fileno(), size) instead of fileobj.read(size).

Even better would be add code to handle buffered file objects. Unfortunately, I think that BufferedIOBase does not expose whether it has data in its buffer.

Start with the selectors documentation and the BufferedIOBase behavior referenced in the issue, then compare examples using sys.stdin and socket.makefile(). Determine whether the scope is a documentation warning or implementation support for buffered data. Done means the chosen behavior is documented clearly, including the os.read(fileobj.fileno(), size) recommendation if the change remains documentation-only.