So it turns out that adding:
[Service]
RuntimeDirectory=php-fpm
RuntimeDirectoryMode=0755
does indeed force the creation of the /var/run/php-fpm folder, but this is needed only on Centos7.
And of course I messed up. I was changing a mix of CentOS8 machines (where the problem does not exist - bottom section explains why) and CentOS7 machines, ran systemctl edit on the CentOS7 machinines, then checked for the existence of /run/php-fpm on a CentOS8 machine (too many terminals).
What happens is, on CentOS8 php-fpm is the default for apache, and the remi www.conf, as well as the supplied apache config php74-php-fpm.conf reflect that, setting the socket folder to /var/opt/remi/php74/run/php-fpm/ (a non volatile folder).
On Centos7, however, mod_php is the default, and remi apache configuration file makes no php-fpm provision (one has to set it up in the site config files). Following the native php-fpm config, I set the socket folder to /run/php-fpm, which is volatile and managed by systemd - making it necessary to change the php74-php-fpm service definition in order to create the folder at startup.
So it appears that there are multiple methods of managing volatile directories, which I think is confusing and error prone, as it appears people are using both with no discernible rationale. But what do I know.