cmd/evm: make batched state-test execution possible

[holiman]

This PR implements a feature that is already present in both nethermind and besu: the ability to run several state-tests in one instance.
For state-tests that are reasonably small, this makes a huge difference. Here's 1000 executions of a state-test which has a run-length of merely 28 steps.

Doing 1000 execs old style

real	0m55.600s
user	0m41.665s
sys	0m17.886s

Doing 1000 execs new style

real	0m1.400s
user	0m1.443s
sys	0m0.116s

Nearly one minute versus <2seconds. When running it in a fuzzing-setting, the speedups is less marked, somewhere in the 1x-2x region.
If the kzg stuff now has a static overhead, as suggested in #27258, this PR would definitely help (the fuzzer). Perhaps something similar can be done for evm t8n too.

myfunc() {
 for i in {1..1000}; do
        ./evm_master --json statetest /home/user/go/src/github.com/holiman/goevmlab/evms/testdata/cases/00000936-mixed-1.json  2>/dev/null 1>/dev/null
 done
}

myfunc2() {
 touch /tmp/foo
 rm /tmp/foo
 for i in {1..1000}; do
        echo "/home/user/go/src/github.com/holiman/goevmlab/evms/testdata/cases/00000936-mixed-1.json" >> /tmp/foo
 done
  echo "" >> /tmp/foo
  cat /tmp/foo | ./evm_batch --json statetest 2>/dev/null 1>/dev/null
}


echo "Doing 1000 execs old style"
time myfunc
echo "Doing 1000 execs new style"
time myfunc2

[MariusVanDerWijden]

LGTM a great feature which will speed up the fuzzing significantly