remove extra stat() call from .pm opening+remove extra safepath check

remove extra stat() call from .pm opening+remove extra safepath check

Originally S_doopen_pm had 2 stat calls, one on the .pm path, and another
on the .pmc, to get mtimes of both. Commit a91233bf4c "Load .pmc always,
even if they are older than a matching .pm file." (see
http://www.nntp.perl.org/group/perl.perl5.porters/2006/03/msg110639.html )
got rid of one of the stat calls but the other was left in place, possibly
as an oversight. S_check_type_and_open itself does another stat call on
unix to check for bad kinds of FS entries (reading a dir as a file), so
assuming someone used .pmc files, a good .pmc would be stat, stat,
open instead of the ideal, stat, open. Remove the extra stat from
S_doopen_pm for efficiency. Since the timestamp compare was removed, the
role of S_doopen_pm has been to verify an attempted path is acceptable to
pass to the FS (no IO done), and generate a .pmc path (no IO done), the
IO side of thing is in S_check_type_and_open, it shouldn't be in
S_doopen_pm.

On Win32, on a no .pmc build, an open is directly done on the attempted
.pm path for efficiency, no stat is done normally (see commit d345f48775
"Win32: stat() only after a failed open() on a module"). Before this patch
the .pmc attempted path got a stat which on Win32 is more than 1 IO call,
compared to Win32 open which is 1 IO call. With this patch, the Win32
specific IO logic in S_check_type_and_open executes instead of a generic
Win32 stat so there is just 1 failing IO call for file not found (typical
case for .pmc) instead of multiple file not found IO calls. See ticket for
details.

When .pmc files are enabled (enabled is default), 2 checks for bad null
char paths were done, once in S_doopen_pm, then again in lower level
S_check_type_and_open. Do the check only once in the higher level call
(S_doopen_pm) for efficiency, there is no way for string "c" which is
catted on to contain a null. There was an existing comment refering to
the problem of a low level check for null returning a message about a
".pmc" instead of a ".pm", so that is another reason to do it at a higher
level. Note on no PMC builds, S_check_type_and_open replaces S_doopen_pm
and still must do the check.