WL#353: Better REGEXP package

We should try to replace the current REGEXP package (Henry Spencer's) with a
newer one because the current one is extremely slow when you search after
multiple strings:

text REGEXP 'tom|david|monty'

Sergei: Not only because it's slow, but in 4.1 the main reason is - because it
only works with simple charsets, no multi-byte, no complex (german) collation rules.

The challenge is to find a compatible but faster REGEXP library that we can
modify to work with our character set handling...

Examples of multi-byte aware regex libraries: pcre, tre.

References:
- BUG#30241 Regular expression problems