Fix ReDoS caused by very large character references using repeated 0s

This patch will fix the ReDoS that is caused by large string of 0s on a character reference (like `&#00000000...`).

## Proof of Concept
```ruby
require "rexml"
require "benchmark"

def test(benchmark, payload)
  benchmark.report { begin REXML::Document.new(payload) rescue Exception end }
end

Benchmark.bm do |x|
  test(x, '<test testing="&#' + "0" * 20000 + '"/>')
  test(x, '<test testing="&#' + "0" * 40000 + '"/>')
  test(x, '<test testing="&#' + "0" * 60000 + '"/>')
  test(x, '<test testing="&#' + "0" * 80000 + '"/>')
  test(x, '<test testing="&#' + "0" * 100000 + '"/>')
end
```

Author: Watson1978

lib/rexml/text.rb

@@ -151,25 +151,37 @@ def Text.check string, pattern, doctype
         end
       end
 
-      # context sensitive
-      string.scan(pattern) do
-        if $1[-1] != ?;
-          raise "Illegal character #{$1.inspect} in raw string #{string.inspect}"
-        elsif $1[0] == ?&
-          if $5 and $5[0] == ?#
-            case ($5[1] == ?x ? $5[2..-1].to_i(16) : $5[1..-1].to_i)
-            when *VALID_CHAR
+      pos = 0
+      while (index = string.index(/<|&/, pos))
+        if string[index] == "<"
+          raise "Illegal character \"#{string[index]}\" in raw string #{string.inspect}"
+        end
+
+        unless (end_index = string.index(/[^\s];/, index + 1))
+          raise "Illegal character \"#{string[index]}\" in raw string #{string.inspect}"
+        end
+
+        value = string[(index + 1)..end_index]
+        if /\s/.match?(value)
+          raise "Illegal character \"#{string[index]}\" in raw string #{string.inspect}"
+        end
+
+        if value[0] == "#"
+          character_reference = value[1..-1]
+
+          unless (/\A(\d+|x[0-9a-fA-F]+)\z/.match?(character_reference))
+            if character_reference[0] == "x" || character_reference[-1] == "x"
+              raise "Illegal character \"#{string[index]}\" in raw string #{string.inspect}"
             else
-              raise "Illegal character #{$1.inspect} in raw string #{string.inspect}"
+              raise "Illegal character #{string.inspect} in raw string #{string.inspect}"
             end
-          # FIXME: below can't work but this needs API change.
-          # elsif @parent and $3 and !SUBSTITUTES.include?($1)
-          #   if !doctype or !doctype.entities.has_key?($3)
-          #     raise "Undeclared entity '#{$1}' in raw string \"#{string}\""
-          #   end
           end
         end
+
+        pos = end_index + 1
       end
+
+      string
     end
 
     def node_type

test/parse/test_character_reference.rb

@@ -0,0 +1,20 @@
+require "test/unit"
+require "core_assertions"
+
+require "rexml/document"
+
+module REXMLTests
+  class TestParseCharacterReference < Test::Unit::TestCase
+    include Test::Unit::CoreAssertions
+
+    def test_gt_linear_performance_malformed_character_reference
+      seq = [10000, 50000, 100000, 150000, 200000]
+      assert_linear_performance(seq, rehearsal: 10) do |n|
+        begin
+          REXML::Document.new('<test testing="&#' + "0" * n + '"/>')
+        rescue
+        end
+      end
+    end
+  end
+end