Optimize ripper translator token sorting

With the benchmark from 2ea8139
Prism is:
* 1.33x slower before
* 1.07x slower after

Author: Earlopain

lib/prism/lex_compat.rb

@@ -802,8 +802,13 @@ def result
       # Drop the EOF token from the list
       tokens = tokens[0...-1]
 
-      # We sort by location to compare against Ripper's output
-      tokens.sort_by!(&:location)
+      # We sort by location because `Ripper.lex` sorts. `Ripper::Lexer.new("foo").parse` emits tokens
+      # in the same order that prism does. Since that method is not part of the public API, its behavior
+      # is not replicated. Manually implemented instead of `sort_by!(&:location)` for performance.
+      tokens.sort_by! do |token|
+        line, column = token.location
+        source.line_to_byte_offset(line) + column
+      end
 
       # Add :on_sp tokens
       tokens = add_on_sp_tokens(tokens, source, result.data_loc, bom, eof_token)