Optimize Font::CMap#read_codes to lower memory usage

The current code use string.each_byte which returns an Enumerator. It
then uses calls to enum.next for getting the next byte.

It turns out that this is quite wasteful in terms of memory usage as
each call to #next seems to allocate (don't really know why) an
Enumerator, an Array, a Proc and a Fiber.

By using manual iteration all these allocations are avoided, leading to
quite noticeable lower memory usage (sample script that decodes text of a
PDF):

Before:
    Initial memory: 17.26MB, Memory maps: 179
    Processing 306 pages...
    <SNIP/>
    Final memory: 193.55MB (total growth: +176.29MB)
    Final memory maps: 14485 (total growth: +14306)

After:
    Initial memory: 17.31MB, Memory maps: 177
    Processing 306 pages...
    <SNIP/>
    Final memory: 73.47MB (total growth: +56.16MB)
    Final memory maps: 200 (total growth: +23)

Author: gettalong

CHANGELOG.md

@@ -7,6 +7,8 @@
 
 ### Changed
 
+* Optimized decoding character codes with a CMap to drastically lower memory
+  usage
 * CLI command `hexapdf inspect rev` to show whether the cross-reference table
   was reconstructed
 

lib/hexapdf/font/cmap.rb

@@ -143,20 +143,24 @@ def add_codespace_range(first, *rest)
       # An error is raised if the string contains invalid bytes.
       def read_codes(string)
         codes = []
-        bytes = string.each_byte
+        bytes = string.bytes
+        length = bytes.length
+        i = 0
 
-        loop do
-          byte = bytes.next
+        while i < length
+          byte = bytes[i]
+          i += 1
           code = 0
 
           found = @codespace_ranges.any? do |first_byte_range, rest_ranges|
             next unless first_byte_range.cover?(byte)
 
             code = (code << 8) + byte
             valid = rest_ranges.all? do |range|
-              begin
-                byte = bytes.next
-              rescue StopIteration
+              if i < length
+                byte = bytes[i]
+                i += 1
+              else
                 raise HexaPDF::Error, "Missing bytes while reading codes via CMap"
               end
               code = (code << 8) + byte