perf(evm): investigate FuzzedExecutor overhead

[Evalir]

We had a performance regression while doing an extra fuzz with 1 run on #5205. While it's understandable that more fuzz runs from scratch will add overhead, the resulting regression is still somewhat surprising given it was only 1 run.

We should investigate the FuzzedExecutor and adjacent things to see if we can squeeze some more performance out of it.

[Evalir]

Closing for lack of appetite—we'll probaby revisit once we get to rewriting the fuzzer (a bit into the distant future though)