This removes the use of HMap in the optimizeWriteBatch function, which is not needed, since we just need a list of writes and to keep them type-aligned.
This may be marginally faster at submission since we don't do the HMap lookup, which is a bit expensive on big TypedPipe graphs (although optimization requires a lot of that).
It may have given a clearer error message in the case of #1851 since we wouldn't have wrongly thought fetching from the list was broken (although, the optimizer would likely have given weird results in any case).
This removes the use of HMap in the optimizeWriteBatch function, which is not needed, since we just need a list of writes and to keep them type-aligned.
This may be marginally faster at submission since we don't do the HMap lookup, which is a bit expensive on big TypedPipe graphs (although optimization requires a lot of that).
It may have given a clearer error message in the case of #1851 since we wouldn't have wrongly thought fetching from the list was broken (although, the optimizer would likely have given weird results in any case).