CSV writer improvements

[rcaudy]

1. Change CSV writing code to use chunk-based reading code and stop allocating boxed primitives.
2. Correct some trivial warnings.
3. Fix column header separator-escaping bug.

[rcaudy]

Unit tests have been updated to provide full coverage for the new/changed code.
Some basic testing suggests significant speedup in cases where chunk filling is more performant than serial get.

[devinrsmith]

I see that using Table.columnIterators would be more expensive because they don't use a shared context and have individual keys; but that makes me wish we had a iterator-like construction that could be as efficient as this.

ColumnsIterator it = table.columnsIterator();
OfByte bytes = it.byteColumn("Foo");
OfDouble doubles = it.doubleColumn("Bar");
while (it.hasNext()) {
  byte myByte = bytes.getByte();
  double myDouble = doubles.getDouble();
  it.advance();
}
it.close()

Having something like this would make it easier IMO to splay out to these sorts of row-oriented formats without having to worry as much about low-level chunking details.

Happy to approve PR otherwise, but had a few Qs.

[rcaudy]

I see that using Table.columnIterators would be more expensive because they don't use a shared context and have individual keys; but that makes me wish we had a iterator-like construction that could be as efficient as this.

ColumnsIterator it = table.columnsIterator();
OfByte bytes = it.byteColumn("Foo");
OfDouble doubles = it.doubleColumn("Bar");
while (it.hasNext()) {
  byte myByte = bytes.getByte();
  double myDouble = doubles.getDouble();
  it.advance();
}
it.close()

Having something like this would make it easier IMO to splay out to these sorts of row-oriented formats without having to worry as much about low-level chunking details.

Happy to approve PR otherwise, but had a few Qs.

That's a clever idea. It might be a lot easier to use for many use cases. It obviously wouldn't be a Java Iterator, though, which makes it less suitable for other use cases.