[PATCH] - Provide robust alternatives for replace_string

Hi,

In our testing framework, backed by pg_regress, there exists the ability to use special strings
that can be replaced by environment based ones. Such an example is '@testtablespace@'. The
function used for this replacement is replace_string which inline replaces these occurrences in
original line. It is documented that the original line buffer should be large enough to accommodate.

However, it is rather possible and easy for subtle errors to occur, especially if there are multiple
occurrences to be replaced in long enough lines. Please find two distinct versions of a possible
solution. One, which is preferred, is using StringInfo though it requires for stringinfo.h to be included
in pg_regress.c. The other patch is more basic and avoids including stringinfo.h. As a reminder
stringinfo became available in the frontend in commit (26aaf97b683d)

Because the original replace_string() is exposed to other users, it is currently left intact.
Also if required, an error can be raised in the original function, in cases that the string is not
long enough to accommodate the replacements.

Worthwhile to mention that currently there are no such issues present in the test suits. It should
not hurt to do a bit better though.

//Asim and Georgios