Memoize subdivision YAML loading

[mdehoog]

Currently the subdivision YAML files are being loaded for every new Country instance. This PR memoizes the loaded YAML file on a class instance variable.

Also replaced the inject -> merge pattern with a more performant version.

Performance changes:
From:

pry(main)> Benchmark.realtime { 10000.times { Country['US'].subdivisions } }
=> 17.20118099998217

To:

pry(main)> Benchmark.realtime { 10000.times { Country['US'].subdivisions } }
=> 0.6878289999440312

[rposborne]

@mdehoog What are the memory benchmarks on this? But this is awesome.

[mdehoog]

Before:

pry(main)> Benchmark.memory { |x| x.report { Country.all.each(&:subdivisions) } }
Calculating -------------------------------------
                        16.663M memsize (     0.000  retained)
                       130.895k objects (     0.000  retained)
                        50.000  strings (     0.000  retained)

After:

pry(main)> Benchmark.memory { |x| x.report { Country.all.each(&:subdivisions) } }
Calculating -------------------------------------
                         8.122M memsize (     4.710M retained)
                       121.002k objects (    64.401k retained)
                        50.000  strings (    50.000  retained)

More memory is retained (~4.7mb), but interestingly not as much is allocated.

[rposborne]

And for clarity these only loaded if you call subdivisions? So no performance change if you are not using subdivisions?

[mdehoog]

Correct, the subdivisions are still loaded lazily, but are now memoized at a class level rather than an instance level.

[rposborne]

Well hell yea, this is great.