Make Base.numerify single-pass and allocation-free per digit

Replace the sub-then-gsub double scan with a single gsub whose block
tracks whether the leading (non-zero) digit has been emitted, and look
digits up in a frozen DIGITS table instead of calling Integer#to_s for
every replaced character.

Benchmark (Ruby 3.4.9, arm64-darwin25, benchmark-ips):

    require 'benchmark/ips'
    require 'faker'

    Benchmark.ips do |x|
      x.config(warmup: 1, time: 2)
      x.report("numerify('###-###-####')") { Faker::Base.numerify('###-###-####') }
      x.report('PhoneNumber.phone_number') { Faker::PhoneNumber.phone_number }
    end

Results:

    main:        numerify('###-###-####')  486.140k (+/- 4.8%) i/s
                 PhoneNumber.phone_number  100.815k (+/-15.3%) i/s
    this commit: numerify('###-###-####')  601.743k (+/- 5.2%) i/s  (~1.24x)
                 PhoneNumber.phone_number  104.568k (+/-14.7%) i/s  (within noise)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Author: connorshea

lib/faker.rb

@@ -58,6 +58,9 @@ class Base
     ULetters = Array('A'..'Z')
     LLetters = Array('a'..'z')
     Letters = ULetters + LLetters
+    # Frozen digit strings, used to avoid allocating a new string for each
+    # random digit generated by #numerify.
+    DIGITS = %w[0 1 2 3 4 5 6 7 8 9].freeze
 
     class << self
       attr_reader :flexible_key
@@ -66,9 +69,15 @@ class << self
 
       ## by default numerify results do not start with a zero
       def numerify(number_string, leading_zero: false)
-        return number_string.gsub('#') { rand(10).to_s } if leading_zero
-
-        number_string.sub('#') { rand(1..9).to_s }.gsub('#') { rand(10).to_s }
+        leading = !leading_zero
+        number_string.gsub('#') do
+          if leading
+            leading = false
+            DIGITS[rand(1..9)]
+          else
+            DIGITS[rand(10)]
+          end
+        end
       end
 
       def letterify(letter_string)