Just to make sure I understand everything correctly:  to fix vfs and refchan 
implementations in the wild would require:

1:  [$vfs open] to properly decode the "mode" argument and pass it on to refchan.

2:  refchan implementations (such as ::vfs::memchan, ::tcl::chan::variable) to 
    take (require?) a mode argument and pass it on to [chan create]


If I'm observing correctly, [$vfs open] should also pay attention to whether 
truncation has been requested (w w+) and consider whether to return ENOENT
for (r r+).  It does not need to handle binary mode or seeking to the end.


I agree that this is a lot of detail for VFS implementations to get right (or 
risk subtle bugs) but I think we should try to get the reference implementations 
right - and well covered in tests - before plumbing magic channels.