BUG: `nullCount` related regression in `getJSONObject`

nullCount and countNonNulls are not lining up. The nullmask after performing getJSONObject appears, perhaps, to be uninitialized.

rapids@tcomer-NVIDIA:~/node/modules/demo/api-server$ node null_count_test.js public/small-graph.json 
0
100
rapids@tcomer-NVIDIA:~/node/modules/demo/api-server$ node null_count_test.js public/large-graph.json 
0
100
rapids@tcomer-NVIDIA:~/node/modules/demo/api-server$ node null_count_test.js public/graphology.json 
1918976
81023
rapids@tcomer-NVIDIA:~/node/modules/demo/api-server$ cat null_count_test.js 
#! node

const file = process.argv[2];

const {Series} = require('@rapidsai/cudf');
const small = Series.readText(file);
const s2 = small.split('"edges":');
const edges2 = s2.gather([1], false);
const E2 = edges2.split('[\n').gather([1], false).split('},\n')
keys = E2.getJSONObject('.key')
console.log(keys.nullCount)
console.log(keys.countNonNulls())

There are no nulls in graphology.json, and the format of the numbers does not appear to change. This isn't trivial to track down.

This nullMask affects the use of gather directly, and most likely other cases. It caused a regression in my #397 work. I'll work on producing a minimal reproducer.

Now that I have my regression fixed (with setNullMask(1, 0)) I'm moving on for now.

Start with null_count_test.js and reproduce the differing nullCount and countNonNulls results for graphology.json. Trace the nullmask behavior through getJSONObject and gather, using the setNullMask workaround and regression context from #397 as clues. Done means the counts agree for the reproducer and the regression is covered by a test.