The bytecode format was updated so the header version constants must be
updated as well.

Unlike `mrb_check_hash_type()` that returns `nil` if the argument is
not a `Hash`, `mrb_ensure_hash_type()` raises a `TypeError` exception.

We assume meta-programming is less used in embedded environments.
We have moved following methods:

 * Kernel module
   global_variables, local_variables, singleton_class,
   instance_variables, instance_variables_defined?, instance_variable_get,
   instance_variable_set, methods, private_methods, public_methods,
   protected_methods, singleton_methods, define_singleton_methods

 * Module class
   class_variables, class_variables_defined?, class_variable_get,
   class_variable_set, remove_class_variable, included_modules,
   instance_methods, remove_method, method_removed, constants

 * Module class methods
   constants, nesting

Note:
Following meta-programming methods are kept in the core:

 * Module class
   alias_method, undef_method, ancestors, const_defined?, const_get,
   const_set, remove_const, method_defined?, define_method

 * Toplevel object
   define_method

`mruby-metaprog` gem is linked by default (specified in default.gembox).
When it is removed, it will save 40KB (stripped:8KB) on x86-64
environment last time I measured.

This problem only appears when `mrb` executed multiple times (i.e. `mirb`)

Use `uint32_t` instead. Theoretically `uint32_t` can overflow as well,
but I think it is sufficient size for embeddable Ruby VM.

This commit also fix #4096.

Merge branch 'node_negate_fix' of https://github.com/ukrainskiysergey/mruby into ukrainskiysergey-node_negate_fix

Those small stack indexes can cause integer overflow.

A byte was too small to hold ensure&rescue stacks indexes.

`printf()` and related functions should not be called when
`MRB_DISABLE_STDIO` is defined.

The type of `s->pc` is now `uint16_t` that can be overflowed easily.
Need more checks.

Implemented by adding `OP_HASHCAT` that merges hashes.

`gcc -O3` raises error on truncation using `snprintf`.

This time we used `Integral` module which is mruby specific.

`"a".sub!("a", "a")` should not return `nil`.

`Hash#compact!` should return `nil` if the receiver does not change.

`mrb_iv_p` -> `mrb_iv_name_sym_p`
`mrb_iv_check` -> `mrb_iv_name_sym_check`

Use segment list for instance variable again to reduce memory.

The macro `RCLASS_SUPER`, `RCLASS_IV_TBL` and `RCLASS_M_TBL` are
removed from `include/mruby/class.h`.