Performance notes

• Replacement with a format message doesn't have great performance:
  • if only 2/20 matches are used, there are 18 useless allocations
• re.nim's find takes 7.74s, nre's find takes 22.35s, python's find takes 25.15s. mostly gc time allocating the match cache.

# Children      Self  Command  Shared Object    
# ........  ........  .......  .................
#
    99.99%    89.09%  test     test             
               |          
               |--12.02%-- pcre_exec
               |          matchimpl_138296
               |          
               |--11.98%-- collectctbody_72219
               |          newseq_135877
               |          match_135928
               |          
               |--5.23%-- newseq_138339
               |          matchimpl_138296
               |          
               |--4.95%-- matchimpl_138296
               |          
               |--4.93%-- nimGCvisit
               |          |          
               |          |--65.53%-- TMP94
               |          |          collectctbody_72219
               |          |          newseq_135877
               |          |          match_135928
               |          |          
               |          |--18.19%-- TMP93
               |          |          collectctbody_72219
               |          |          newseq_135877
               |          |          match_135928
               |          |          
               |           --16.28%-- collectctbody_72219
               |                     newseq_135877
               |                     match_135928