Minor `I18n.normalize_keys` improvement

[codealchemy]

Minor performance bump, as well as fewer objects allocated, when using * v. concat for normalizing keys -

📝 Last updated in https://github.com/ruby-i18n/i18n/commit/6425f554c2f5ea5c22365632d818f9f04239023d

Measurements

```ruby # Ruby 2.7.5 require 'benchmark/ips' report = Benchmark.ips do |x| x.report("concat") do ex = [] ex.concat ['foo'] ex.concat ['bar', 'baz'] ex.concat ['quix'] ex end x.report("splat") do [ *['foo'], *['bar', 'baz'], *['quix'] ] end x.compare! Warming up -------------------------------------- concat 185.470k i/100ms splat 467.395k i/100ms Calculating ------------------------------------- concat 2.176M (± 1.5%) i/s - 10.943M in 5.030233s splat 4.678M (± 2.0%) i/s - 23.837M in 5.098138s Comparison: splat: 4677698.6 i/s concat: 2175917.9 i/s - 2.15x (± 0.00) slower ``` ```ruby require 'memory_profiler' n = 100 report1 = MemoryProfiler.report do n.times do ex = [] ex.concat ['foo'] ex.concat ['bar', 'baz'] ex.concat ['quix'] ex end end report1.pretty_print(scale_bytes: true, detailed_report: false) # Total allocated: 48.00 kB (800 objects) # Total retained: 0 B (0 objects) report2 = MemoryProfiler.report do n.times do [ *['foo'], *['bar', 'baz'], *['quix'] ] end end report2.pretty_print(scale_bytes: true, detailed_report: false) # Total allocated: 23.20 kB (500 objects) # Total retained: 0 B (0 objects) ```

[radar]

Thank you!