Remove `.gitkeep` files

Remove special treatments for `MRB_WITHOUT_FLOAT` in build scripts

Avoid runtime evaluation for `MRB_WITHOUT_FLOAT`

Fix assertion name for `Numeric#**` test

Fix coercing for first step counter in `Numeric#step`

Before:

  a=[]; 7.step(4, -3.0) { |c| a << c }; p a  #=> [7, 4.0]

After / Ruby:

  a=[]; 7.step(4, -3.0) { |c| a << c }; p a  #=> [7.0, 4.0]

[ci skip] Sort `.gitignore`

Build fails when set `conf.build_dir=<rel path>` and `conf.enable_cxx_exception`

Ignore compiler temporary files (for `-save-temps` flag)

Use `%g` instead of `%e` for float representation in dump format

Improve compatibility to CRuby for `Float#to_s`

For conforming C standard.

Bfore:

  Float::INFINITY.to_s  #=> "inf"
  50.0.to_s             #=> "50"
  1e20.to_s             #=> "1e+20"

After / CRuby:

  Float::INFINITY.to_s  #=> "Infinity"
  50.0.to_s             #=> "50.0"
  1e20.to_s             #=> "1.0e+20"

Use `__send__` instead of `send`; ref #4207

`%g` use shorter representation than `%e`.

Remove duplicate code in numeric.c

Change the order of "expected" and "actual" in test

Fix dump/load float leteral evaluate to infinity

Add `Time` object creation API

Example:

  # example.rb
  p(2e308)
  p(-2e308)

  Good:

    $ bin/mruby example.rb
    inf
    -inf

  Bad:

    $ bin/mrbc example.rb
    $ bin/mruby -b example.mrb
    0
    -0

Cause:

  Float infinity representation is `inf` on dump and it is converted by
  corresponding `String#to_f` on load.

Treatment:

  - Introduce new representations (`i`: +infinity, `I`: -infinity)
  - Allow old representations (`inf`, `-inf`, `infinity`, `-infinity`) too
  - Raise error for unknown representations (use corresponding `Kernel#Float`)

Fix `0.0` and `-0.0` handling.

Fix the following issue:

    Good:
    $ bin/mruby -e 'p(-0.0)' #=> "-0"

    Bad:
    $ bin/mruby -e 'a=0.0; p(-0.0)' #=> "0"

Add `assert_same` and `assert_not_same`

Integrate mrblib/float.rb into src/numeric.c