Add Multijoin feature to community

[lbooker42]

Port the enterprise multijoin feature to allow more efficient computation of multiple table joins.

The multiJoin operation collects the set of distinct keys from the input tables, and then joins one row from each of the input tables onto the result. Input tables need not have a match row for each key, but they may not have multiple matching rows for a given key. The operation can be thought of as a merge of the key columns, followed by a selectDistinct and then a series of iterative naturalJoin operations as follows:

private Table doIterativeMultiJoin(String [] keyColumns, List<? extends Table> inputTables) {
    final List<Table> keyTables = inputTables.stream().map(t -> t.view(keyColumns)).collect(Collectors.toList());
    final Table base = TableTools.merge(keyTables).selectDistinct(keyColumns);

    Table result = base;
    for (int ii = 0; ii < inputTables.size(); ++ii) {
        result = result.naturalJoin(inputTables.get(ii), Arrays.asList(keyColumns));
    }

    return result;
}

[rcaudy]

Review of just the changes since Tuesday.

[rcaudy]

WIP on review

[rcaudy]

Review should continue with MultiJoinMergedListener and the actual state managers.

[rcaudy]

Not for this PR, but maybe this should look like NanosBasedTimeArraySource, and be a wrapper around a ByteArraySource.

[lbooker42]

Created #4272 to track this request.

[rcaudy]

WIP. Only incremental state managers left.

[rcaudy]

Reviewed current state of incremental state managers.

[rcaudy]

We could probably memoize this, using the memoization cache on the first table.

[rcaudy]

This can hold for another PR.

[lbooker42]

Created issue #4273 to track this.

[rcaudy]

Does this assignment effect anything? I think it's immaterial, since we'll ignore any hashed values that are greater than the rehashPointer (which is zero). Some part of me wants to assign it to oldTableSize in newAlternate, and back to 1 in clearAlternate, just for cleanliness.

[lbooker42]

oldTableSize is local to doRehash() and the newAlternate() signature is determined by the core hash generation code. Adding the old size as a parameter is a significant change that will affect all existing hashing routines.

The alternative is to push oldTableSize into a member variable so it can be accessed by newAlternate(). Do you want me to do this?

[rcaudy]

I think what you have done is good.

[rcaudy]

Although I don't love that you're clearing it in clearAlternate, and assigning it after newAlternate, I think this is good enough for now.

[deephaven-internal]

Labels indicate documentation is required. Issues for documentation have been opened:

How-to: https://github.com/deephaven/deephaven.io/issues/2995
Reference: https://github.com/deephaven/deephaven.io/issues/2996