1) Port NetBSD VMIN/VTIME stuff:

1) Port NetBSD VMIN/VTIME stuff:
	fix bug in case VMIN=0 && rawq len == 0,
	fix bug with missing untimeouts.

2) Fixes from bde:
	o Don't export private functions.
	o Don't declare externs inside functions.  I moved the declaration
	  of `nldisp' to the top of tty.c; it should be declared in some
	  header.
	o Moved ttywait() deadlock fix from sio.c to here so it doesn't
	  harm efficiency and so it works with other drivers.
	o Use "cuddle else style" ( "} else"  not "}\n\t..else" ).
	o Don't clear TS_TBLOCK before it is used in ttyflush() (your version
	  had this from a previous incomplete fix).
	o Added long comment about probable botch of TS_ISOPEN and TS_WOPEN
	  flags when CLOCAL drops while there is no carrier.
	o Added comment about races in ttnread().  Only some callers use
	  spltty().  ttselect() doesn't.
	o Fixed use of random types (int, long) instead of tcflags_t and
	  use of u_char instead of cc_t (you missed the changes in my
	  previous diffs).
	o "int alt = lflag&ALTWERASE" had the wrong type.  It's only used
	  once, so don't use a flag variable for it.
	o Fixed (? :-) use of MIN() and min() on integral types.  MIN()
	  expands to min() and min() is an (inline) function that really
	  only works for u_int types.  "gcc -Wconversion" told me about
	  the sign botches.  Of course, a min of -1 would be fatal as
	  the count arg for bcopy or bzero.
	o Fixed missing loop exit after uiomove() failure in ttwrite()
	  (bug introduced in my previous version :-().
	o Fixed serious races in ttwrite().  The old clist buffer access
	  functions apparently called spltty()/splx() (NetBSD's getc()
	  and putc() do it) although this makes them slow.  Lots of
	  races were introduced by the non-locking RB_* access macros.
	o Cleaned up fix for ECHONL again.
	o tp->t_gen is now int; use int for `gen' copy of it too.