This commit basically implements two approaches
to improve the performance of RubyHash:

Switching from closed addressing hashing (double linked list)
to open addressing hashing because of better cache locality.
Furthermore we removed almost all RubyHashEntry objects
for smaller memory allocation (which helps for better cache
locality as well).

Further more small hashes (less than 8 entires) are now
implemented via a linear search which reduces memory
allocation for a buckets. For fast bucket skips we maintain
in this case a hashes array to cache the hash values.

Implements #4708 & #2989