Improve performance of Oj.dump with compat/rails mode

[Watson1978]

This patch introduces oj_hash_has_key() (same as rb_hash_has_key)
to reduce rb_funcall() calling because it has a overhead.

This patch will improve Oj.dump performance as following.

−	before	after	result
Oj.dump	1.949M	1.966M	−
Oj.dump (compat)	850.154k	1.198M	1.41x
Oj.dump (rails)	657.383k	840.051k	1.28x

Environment

• MacBook Air (M1, 2020)
• macOS 12.0 beta 3
• Apple M1
• Ruby 3.0.2

Before

Warming up --------------------------------------
             Oj.dump   198.379k i/100ms
    Oj.dump (compat)    86.466k i/100ms
     Oj.dump (rails)    66.760k i/100ms
Calculating -------------------------------------
             Oj.dump      1.949M (± 0.3%) i/s -      9.919M in   5.088487s
    Oj.dump (compat)    850.154k (± 0.3%) i/s -      4.323M in   5.085367s
     Oj.dump (rails)    657.383k (± 0.4%) i/s -      3.338M in   5.077802s

After

Warming up --------------------------------------
             Oj.dump   198.297k i/100ms
    Oj.dump (compat)   120.402k i/100ms
     Oj.dump (rails)    84.204k i/100ms
Calculating -------------------------------------
             Oj.dump      1.966M (± 0.3%) i/s -      9.915M in   5.044305s
    Oj.dump (compat)      1.198M (± 0.2%) i/s -      6.020M in   5.026524s
     Oj.dump (rails)    840.051k (± 0.1%) i/s -      4.210M in   5.011848s

Test code

require 'benchmark/ips'
require 'oj'

data = {
  'short_string': 'a' * 50,
  'long_string': 'b' * 255,
  'utf8_string': 'あいうえお' * 10
}

Benchmark.ips do |x|
  x.report('Oj.dump') { Oj.dump(data) }
  x.report('Oj.dump (compat)') { Oj.dump(data, mode: :compat) }
  x.report('Oj.dump (rails)') { Oj.dump(data, mode: :rails) }
end

[Watson1978]

I retrieved a benchmark result on Intel Mac.

–	before	after	result
Oj.dump	1.435M	1.453M	–
Oj.dump (compat)	558.205k	897.370k	1.61x
Oj.dump (rails)	428.819k	655.485k	1.53x

Env

• MacBook Pro (16-inch, 2019)
• macOS BigSur 11.5
• CPU 2.4 GHz 8cores Intel Core i9
• Ruby 3.0.2

Before

Warming up --------------------------------------
             Oj.dump   139.896k i/100ms
    Oj.dump (compat)    55.788k i/100ms
     Oj.dump (rails)    43.009k i/100ms
Calculating -------------------------------------
             Oj.dump      1.435M (± 2.3%) i/s -      7.275M in   5.072783s
    Oj.dump (compat)    558.205k (± 2.5%) i/s -      2.789M in   5.000364s
     Oj.dump (rails)    428.819k (± 4.7%) i/s -      2.150M in   5.026970s

After

Warming up --------------------------------------
             Oj.dump   142.370k i/100ms
    Oj.dump (compat)    89.373k i/100ms
     Oj.dump (rails)    66.401k i/100ms
Calculating -------------------------------------
             Oj.dump      1.453M (± 3.4%) i/s -      7.261M in   5.001952s
    Oj.dump (compat)    897.370k (± 1.9%) i/s -      4.558M in   5.081196s
     Oj.dump (rails)    655.485k (± 2.3%) i/s -      3.320M in   5.067834s

[Watson1978]

@ohler55 I would appreciate any feedback about this pull request. Thanks.

[ohler55]

Looks very reasonable to me. Thank you.

[ohler55]

I would not expect a significant change on non-trivial objects. For small it does help.