Improved matching performance for big datasets

Many changes:

• Probabalistic matching now on a field-by-field basis rather than lumping all tokens into one
• Search algorithm can now account for scores of matches found, stopping when a good match is detected
• Better scoring algorithm that accounts for tokens which don't match and misspellings
• Memoisation to improve performance