Add erroroffset support to perltest.sh and fix discrepancies

[NWilson]

• perltest.sh would need to support outputting error offsets from the Perl messages
• RunPerlTest would need to perform a "loose" diff on certain lines, to handle the fact that the Perl error messages won't have matching error codes
• Migrate as many error cases as possible from testinput{2,5} → testinput{1,4}
• Fix as many discrepancies as possible between the error offsets reported by Perl and PCRE2

There's no particular urgency to this. It would be nice though, perhaps when I have spare time and all the other more urgent tasks on my list are done.

[NWilson]

I have made quite a few fixes. The three remaining cases are:

• PCRE2_ERROR_BAD_SUBPATTERN_REFERENCE
• PCRE2_ERROR_DEFINE_TOO_MANY_BRANCHES
• PCRE2_ERROR_BACKSLASH_K_IN_LOOKAROUND

There's a code TODO for each. These cases basically return a bad or unhelpful error-offset.

When they are fixed, I would consider extracting the pcre2test function error_direction() into a new public helper. It would indicate, for a given error value, whether the syntax issue is to the left or to the right of the error-offset.

The new function can be of the form pcre2_get_error_offset_direction(compile_context* ctx), taking the compile context and returning additional error information.

For the use-cases we have, I have considered and basically ruled out converting the single error offset into a start/end span. That would be laborious, and we don't even know if anyone would ever use it. However, we would have the option in future of adding pcre2_get_error_offset_length(compile_context* ctx) to indicate the length of the span.

It basically comes down to "how many customers care about offering really nice syntax highlighting to pinpoint where the error is". For now, I think offering a good, high-quality single offset (with left/right direction) is good enough.