doc: document limitations with PCRE2_MATCH_INVALID_UTF

[carenas]

Closes: #187

[PhilipHazel]

I agree there should be some rewording, but I'm not sure it should mention breaking down 16-bit units. PCRE expects 16-bit data to be in uint16_t vectors, that is, in the natural BE/LE arrangement for the environment. If somebody is handling 16-bit data in bytes, it is their responsibility to get it into the right format for PCRE. I will work on the documentation to try to make this clear.

[carenas]

The proposed change doesn't mention breaking down 16-bit units. The misguided attempt to suggest doing so as a workaround was only part of the original version, because it was explicitly mentioned in the ticket, and only was added because I found it funny while testing as it would match an UTF-16 'a' out of pure luck when using an BE 'a' after half a broken other composite character:

code = pcre2_compile(L'a', 1, PCRE2_MATCH_INVALID_UTF, &error, &erroff, NULL);
..
int buf = L'a';
buf <<= 8;
buf |= 0xd8000000;
memcpy(subject, &buf, sizeof(int));
error = pcre2_match(code, (char *)subject + 1, 3, 0, 0, data, NULL);
assert(error == 1);

[PhilipHazel]

It took me a while to understand exactly what this was all about. I have added some slightly different wording in the places you suggest, in particular making it clear that this applies only to the 16-bit and 32-bit libraries.