If out of memory occurs with mrb_open(), the process will be crashed by SIGABRT or SIGSEGV.

This behavior is undesirable for me as I try to generate multiple mrb_state.

This patch set corrects problems that occurred by simulating out of memory.
(branch is in fix-mrb_open-with-nomem)

However, it is divided into 18 steps commit, and some can be made independent.

Should it be better to divide the pull requests for each, or may they be a single pull request at once?

Thanks.

Potential issue

If mrb_XXX_gem_final() is called while initialization processing is not completed, problems may arise there.