Implement open addressing algorithm for RubyHash

to improve the performance by leverage better
cache locality.

Switching from closed addressing hash algorithm (linked list)
to open addressing hashing because of a better cache locality
on modern CPU architectures.
Furthermore we removed almost all RubyHashEntry objects
for smaller memory allocation.
This is already implemented in MRI since 2.4, see
https://bugs.ruby-lang.org/issues/12142

Small hashes (less than 8 entries) are now
implemented via a linear search which reduces memory
allocation in this case and has almost no performance
implication. For a fast bucket skip we maintain
in this case a hashes array to cache the hash values.

Implements jruby#4708 & jruby#2989