Any secret escape sequence code that only escape forward slashes?

Don't try to do it all at once. Write a simple function that escape forward slashes, and call that in the replacement part of your substitution. You can use s{}{}ge to evaluated the replacement part, and comfortably call functions there.

#!/usr/bin/perl -w
use strict;

my $line = q{  split( '/',   $string )  };
               
$line =~ s/ //g;  #compress spaces
                  #now just "split('/',$string)"

if ($line =~ m/^split/) # check for the split keyword
{
   # get (parm1, parm2) of the split, ie the two
   # things separated by commas within the parens
   my ($parm1,$parm2) = ($line =~ m/\((.*?),(.*?)\)/)[0,1];
   
   # now get stuff between quotes in parm 1
   my $inside_quote = ($parm1 =~ m|'(.*?)'|)[0];
   
   # change any / to \/
   $inside_quote =~ s|/|\\/|g;
   
   #now just print back out
   print "preg_split('/$inside_quote/',$parm2)\n";
}

__END__
Prints:
preg_split('/\//',$string)
[download]

What's with "\s*?"? Don't you simply mean "\s*"?

Anyway,

s/
    (?<!_)
    ( split \s* \( \s* )
    (
        " (?:[^\\"]+|\\.)* "
    |
        ' (?:[^\\']+|\\.)* '
    )
/
    my @x = ($1,$2);
    $x[1] =~ s{/}{\\/}g;
    "preg_$x[0]$x[1]"
/xesg
[download]

Untested.

Will create an error if the "/" is already escaped.

‘Dumb’ search-and-replaces (and I mean that as a comment on the code, not on you!) always strike fear in my heart; if we can't even parse such a rigid language as XML with regexes (which we can't, right? Or at least no sane person would?), how can we expect correctly to parse the rich grammar of a programming language? I always think of clbuttic.

If this were my job, the first thing I'd do would be to look at some means of getting at the internal, not textual, representation of a PHP program. The first result for PHP + AST is php-ast; I'm not sure if it does what you want, but you might be able to fold, spindle, and mutilate it, or else just look further down in the results.

Dumb search-and-replaces always strike fear in my heart

I do dumb search-and-replaces a hundred times a day, mostly with my editor's search and replace box. There's nothing wrong with it if the process is supervised.

In the OP's case, he can use a visual diff tool — I love Beyond Compare — to compare the pre- and post-change versions of his files and fix the mistakes.

And then he has to do a manual search to change the ones the tool didn't catch.