Two minor correctness bugs found by Clang Analyzer

[NWilson]

I have a PR open which adds Clang's very good static analyzer to our reports.

This has flagged two issues, one of which appears to be a genuine bug:

• In pcre2grep, the end_of_line function is naughty! It looks at the leading byte of a UTF-8 sequence to find out how many bytes to read, then reads them without checking whether the buffer contains enough bytes! It appears to be a read-past-the-end bug if the input ends in a truncated UTF-8 sequence.
• In pcre2posix, the pcre2_regerror has a NULL check for preg, and the function does support passing in a NULL preg. But further down in the function, it's used unconditionally in message_len(message, (int)preg->re_erroffset), and clang reckons that a NULL dereference (might) be possible there. I believe that this is bogus, however, and clang just isn't quite clever enough to work it out; but someone should probably double-check that reasoning.