This function handles zero division properly. Also fixed bugs that multiply
numbers instead of division.

- define `MRB_TT_RATIONAL`
- change object structure (`struct RRational`)
- add memory management for `MRB_TT_RATIONAL`
- avoid operator overloading as much as possible
- implement division overloading in C
- as a result, performance improved a lot

This reverts commit dc51d89a.

Instead of including `mruby/presym.h` everywhere, we provided the
fallback `mruby/presym.inc` under `include/mruby` directory, and specify
`-I<build-dir>/include` before `-I<top-dir>/include` in `presym.rake`.
So even when someone drops `-I<build-dir>/include` in compiler options,
it just compiles without failure.

Addressed an issue where existing programs linking `libmruby.a` could only
be built by adding `<build-dir>/include` to compiler's include path.

`1L` on Windows means `32 bit int`.

Lint

- Implement `Rational()` in `C`.
- Use `float` to `rational` conversion function taken from:
  https://rosettacode.org/wiki/Convert_decimal_number_to_rational#C

Where fixnum overflow can happen.

- Integrate `Fixnum` and `Integer`
- Remove `Integral`
- `int / int -> int`
- Replace `mrb_fixnum()` to `mrb_int()`
- Replace `mrb_fixnum_value()` to `mrb_int_value()`.
- Use `mrb_integer_p()` instead of `mrb_fixnum_p()`

* The `Fixnum` constant is now an alias for the `Integer` class.
* Remove `struct mrb_state::fixnum_class` member.
  If necessary, use `struct mrb_state::integer_class` instead.

- `MRB_WITHOUT_FLOAT` => `MRB_NO_FLOAT`
- `MRB_USE_FLOAT` => `MRB_USE_FLOAT32`

The former is to use `USE_XXX` naming convention. The latter is to make
sure `float` is 32bit float and not floating point number in general.