Yieldable matching API

[NWilson]

The Erlang regex support is based on PCRE2, and they have patched their copy of PCRE2 to allow matching to yield after a fixed amount of work is done, and then resume it later. This enables them to keep the main thread responsive and allocate time fairly to cooperatively-multitasking routines.

This is a noble goal. We should incorporate their support, ensuring that we make the API adequately general-purpose for other users. We also must test it thoroughly before it can be an official upstream feature.

See discussion: https://groups.google.com/g/pcre2-dev/c/vdc2bSvRUuA

Erlang's fork of PCRE2, with changes on a branch: https://github.com/frazze-jobb/pcre2

[frazze-jobb]

Here is the latest Erlang yield branch: https://github.com/frazze-jobb/pcre2/commits/ERLANG_INTEGRATION_pcre2-10.45/

[zherczeg]

How much is the performance overhead when used / not used? Looks like a rather complex code.

Usually software run user code in a separate, heavily right limited process, which can be killed anytime. This way it does not block the main process, and does not need any checkers as well.