Reduce memory usage when parsing lots of different inputs.

Hi,

I am using antlr cppruntime to build a fuzzer, which needs to parse lots of different inputs of a grammar. However, the memory usage is too high (over 50G) after parsing millions of inputs. I can confirm that there are no memory leakage on my parts. I profile the memory usage and they are mostly in `ParserATNSimulator`.

I checked the issues (https://github.com/antlr/antlr4/issues/2182, https://github.com/antlr/antlr4/issues/499), which don't solve my problems: The memory usage is still high and clearing the cache makes the parser really slow sometime (completely stuck for a few minutes).

Every time I parse a input, I create a new parser as below:
```cpp
ANTLRInputStream input(input_program);
XXXLexer lexer(&input);
CommonTokenStream tokens(&lexer);
tokens.fill();
XXXParser parser(&tokens);
```

Are there any ways to limit the memory usage? Any suggestions are appreciated!

Start by profiling ParserATNSimulator while reproducing the C++ parsing loop shown in the issue across many inputs. Review issues 2182 and 499 to understand existing cache-clearing approaches. Done means identifying a concrete way to bound memory growth while preserving acceptable parsing performance, then validating it on the reported workload.

Written by the indexing model from the issue text.