Correctly rewrite bounded repetitions following an explicit class

[masklinn]

Character classes were never unstacked so a bounded repetition following a class would never be rewritten.

This yields no time savings on the bench run, maximum RSS and peak memory do seem to go down by a few MiB (we're talking 129 to 126 or so) but it's not 100% reliable.

[masklinn]

Update: this actually cuts down memory use by the bench significantly but I forgot to recompile the bench example between branch switches 🤦

Before this PR (on a 16" MBP, M1 Pro):

> /usr/bin/time -l ../target/release/examples/bench -r 10 ~/sources/thirdparty/uap-core/regexes.yaml ~/sources/thirdparty/uap-python/samples/useragents.txt
Lines: 751580
Total time: 8.322629583s
11µs / line
        8.75 real         8.49 user         0.03 sys
           136527872  maximum resident set size
                   0  average shared memory size
                   0  average unshared data size
                   0  average unshared stack size
                8546  page reclaims
                   0  page faults
                   0  swaps
                   0  block input operations
                   0  block output operations
                   0  messages sent
                   0  messages received
                   0  signals received
                   0  voluntary context switches
                  63  involuntary context switches
         77382503510  instructions retired
         27456330696  cycles elapsed
           132777216  peak memory footprint

after:

> /usr/bin/time -l ../target/release/examples/bench -r 10 ~/sources/thirdparty/uap-core/regexes.yaml ~/sources/thirdparty/uap-python/samples/useragents.txt
Lines: 751580
Total time: 8.119800417s
10µs / line
        8.48 real         8.20 user         0.03 sys
            57950208  maximum resident set size
                   0  average shared memory size
                   0  average unshared data size
                   0  average unshared stack size
                3750  page reclaims
                   0  page faults
                   0  swaps
                   0  block input operations
                   0  block output operations
                   0  messages sent
                   0  messages received
                   0  signals received
                   0  voluntary context switches
                  68  involuntary context switches
         74965288660  instructions retired
         26498238780  cycles elapsed
            54674752  peak memory footprint

So yeah, #2 should have cut the memory use by 75% or so, which probably puts us within a stone's throw of re2 really.