svm: modularize integration test runner

[2501babe]

Problem

the way the integration test runner is structured, the bank and batch processor are constructed, execute a test entry, run checks as defined by the entry, and then nothing is returned. this makes it difficult to do more bespoke things, like test inspected accounts or look into the final program cache, which we may not want to support in SvmTestEntry

Summary of Changes

introduce SvmTestEnvironment and split execute_test_entry into create and execute. update mock bank state after a batch finishes executing. this allows more complex tests to test things on the test environment after execution, and even to run multiple batches with the same environment

note that, despite the large diff, no code actually changed with the test runner (the first commit) other than the addition of this at the end of execute:

        let mut mock_bank_accounts = self.mock_bank.account_shared_data.write().unwrap();
        *mock_bank_accounts = final_accounts_actual;

otherwise the functions create and execute in SvmTestEnvironment are line-for-line identical with the previous execute_test_entry

the second commit rewrites the svm_inspect_account test to use this new pattern and should be reviewed more closely, although no functionality has changed other than using the actual system program instead of the bpf transfer program, which added nothing to the test. the new pattern means that svm_inspect_account now validates transaction execution status and all final account states implicitly, in addition to the explicit inspected accounts checks

[brooksprumo]

The refactor/modularization looks correct to me. Thanks!