Implement rb_hash_bulk_insert

[Th3-M4jor]

Closes #3705

In this case I'm not sure if the overhead of repeated calls to []= makes it worth instead adding another method to CExtNodes.java

[andrykonchin]

I believe it's OK to have this overhead.

[andrykonchin]

Good job!

[andrykonchin]

I would also add the following cases:

• length isn't even - there should be error raised (I suppose)
• how duplicated keys (provided in the array) are handled - I suppose they should just override each other

 * What happens for  duplicated keys?  Well it silently discards  older ones to
 * accept the newest (rightmost) one.  This behaviour also mimics repeated call
 * of rb_hash_aset().

• it's allowed to pass NULL as array if length argument = 0 (as far as it's explicitly stated in a documentation comment

 * @note        `argv` is allowed to be NULL as long as `argc` is zero.

[Th3-M4jor]

I'll add tests for the other cases.

As for when the length isn't even, it appears to be that if RUBY_DEBUG is not defined MRI accepts an array with an odd number of elements and will read one past the end. Should we allow the same to maintain compatibility?

[andrykonchin]

Ah, right. Yes, it makes sense to do the same and to allow passing odd length.

[eregon]

We shouldn't add a spec for that odd case, because specs should also pass with RUBY_DEBUG, notably ruby-dev-builder depends on that.
@andrykonchin Could you add a commit removing that spec?

We could file an issue at https://bugs.ruby-lang.org/, I think it would make sense, it doesn't seem OK the C API reads past the length it's given.

[andrykonchin]

Looks good! Thank you.