no error handling of key_schedule_new failure

`key_schedule_new` is called from two places, but neither one checks for NULL return value.

As a result, failure to create a hash alg results in a crash (seg fault) happening later on in an unrelated spot.

Can we add something like this after calling `key_schedule_new` in these two spots?
```
if (tls->key_schedule == NULL) {
ret = PTLS_ERROR_NO_MEMORY;
goto Exit;
}
```

Search the repository for the two calls to key_schedule_new and inspect how each caller handles errors and exits. Add the requested NULL-failure handling in both locations, then verify that allocation failure returns PTLS_ERROR_NO_MEMORY rather than causing a later crash.

Written by the indexing model from the issue text.