More work-in-progress on sysread() abstraction.

More work-in-progress on sysread() abstraction.
- sysread() does not want to retry on EINTR but too many places have that logic.
  now stdio and unix are both doing re-try on EINTR it makes sense to move
         that logic up to PerlIO_read()/PerlIO_write() and leave layers as raw without
         retry (which should help them be re-used for sysread).
       - also noted (old?) comment in pp_sys.c that says PerlIO_read() returns 0 on error
  but perlapio.pod says it returns -ve. Match the pod for :stdio.
       - all tests pass (:perlio/:stdio) if pp_sys.c is re-hacked to turn off new-sysread
         logic so above re-factoring is not a disaster...

p4raw-id: //depot/perlio@19617