refactor: ossl verify cb logic

[jmayclin]

The default X509_STORE_CTX only allows for a single verify callback, but s2n-tls 
sometimes requires multiple pieces of functionality for that callback. Previously we 
did this by chaining function together, but this will become more difficult as we add 
more verify functions that we wish to call.

This commit adds a generic config and callback that allows for verify callbacks to 
be more easily configured.

Description of changes:

This commit adds a new s2n_ossl_verify_cb_config struct on the s2n_x509_validator, which is used to control which verify callbacks are invoked.

This is being done in preparation for the cert key restrictions, which will add yet another verify callback to enforce cert key restrictions.

Call-outs:

Alternate implementation.

Rather than adding a new config, we could instead pass the entire s2n_connection * as the application data, and then use the original checks from s2n_x509_validator_verify_cert_chain to toggle the callbacks,
E.g. instead of

    if (callback_config->disable_time_validation
            && s2n_disable_time_validation_ossl_verify_callback(default_ossl_ret, ctx) == OSSL_VERIFY_CALLBACK_OK) {
        return OSSL_VERIFY_CALLBACK_OK;
    }

We could do

    if (conn->config->crl_lookup_cb
            && s2n_disable_time_validation_ossl_verify_callback(default_ossl_ret, ctx) == OSSL_VERIFY_CALLBACK_OK) {
        return OSSL_VERIFY_CALLBACK_OK;
    }

But duplicating all of the conditional checks felt odd, so I went with the config route.

Testing:

This refactor should not change any behavior, which is confirmed by existing unit tests passing.

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.

[lrstewart]

Do we have to set an error code of some kind on failure? Or do we just set the return value?

[jmayclin]

We just set the return value.

[goatgoose]

It might be good to check ctx before giving it to the libcrypto in X509_STORE_CTX_get_app_data().

[jmayclin]

My feature ended up not needing this, and this increases the number of verify cb features we'd be depending on, which is something to avoid if we don't need it.