Fix hash agg

[yuzefovich]

colexec: deeply reset datum-backed vectors in ResetInternalBatch

Previously, we would keep references to the old datums set in the
datum-backed vectors until the datums are overwritten, thus, extending
the time period when the no-longer-used datums are live unnecessarily.
We don't have to do that for any other types because for others we can
actually reuse the old space for new elements, so we want to keep them
around (e.g. decimals can reuse the non-inlined coefficient). This
commit makes it so that we deeply unset the datums in
ResetInternalBatch.

Care had to be taken to keep the memory accounting up-to-date. In
particular, after deeply resetting the datum-backed vector, we want to
release the memory allocation that was taken up by the actual datums
while keeping the overhead of tree.Datum interface in (since
[]tree.Datum slice is still fully live).

Release note: None

colexec: fix limiting the output batch in size in several operators

In the cFetcher, the hash aggregator, and the ordered synchronizer we
are performing careful memory accounting in order to limit the size of
the output batches.

However, we had an incorrect assumption that could lead to batches
being larger than desired. Previously, the first time the batch exceeded
the target size, we would remember its "capacity" (i.e. the number of
rows that fit into that batch), and in the future we would always put up
to that number of rows in the batch. That works well when each row is of
roughly the same size throughout the lifetime of an operator; however,
that is not always the case. Imagine we have 1000 small rows followed by
1000 large rows - previously, all 1000 large rows would be put into
a single batch, significantly exceeding the target size.

This commit makes the limiting much more sane - it removes the notion of
"max capacity" and, instead, after each row is set in the batch we check
whether the batch has exceeded the target size.

Release note: None

[cockroach-teamcity]

This change is