Exponential running time when chaining bounds

[Sjlver]

Expressions like the following cause TRE's run-time to be very high:

echo 'x' | agrep 'x?{100}{100}'
echo 'x' | agrep 'x?{5}{5}{5}{5}{5}{5}'

Found using LLVM's LibFuzzer.

[Sjlver]

I think this can be fixed by allowing only a single repetition operator after every atom. This is what the grammar specifies. However, the implementation currently allows repetiton because it calls STACK_PUSHX(stack, int, PARSE_POSTFIX) again after parsing a postfix operator.

[Sjlver]

My suggested fix does not fully solve the problem. Expressions like (({81}){81}){81} still lead to stack overflows.