Imperfect rollup in native ingestion when using integer dimensions and replacing nulls with default value

### Affected Version
0.18.0

### Description
During native ingestion `RollupFactsHolder` structure is used for maintaining a sorted list of intermediate rows. It treats missing integer values as nulls and puts them before any other value. Although later when an incremental persist is created, by default null values get replaced with 0. This breaks the sorted ordering of rows which prevents merging of identical rows as the next stage algorithm in `RowCombiningTimeAndDimsIterator` relies on incremental persists being sorted when performing the merge.

### Steps To Reproduce
Import two files below with the given ingestion spec (replacing ``).
The expected result is 3 rows total, although 4 rows are produced.

Input file 1:
```
{"time": 1589512112, "dim": -101, "value": 1}
{"time": 1589512112, "dim": -100, "value": 2}
```
Input file 2:
```
{"time": 1589512112, "dim": null, "value": 10}
{"time": 1589512112, "dim": -100, "value": 20}
```
Ingestion spec:
```
{
"type": "index_parallel",
"spec": {
"ioConfig": {
"type": "index_parallel",
"inputSource": {
"type": "local",
"filter": "*",
"baseDir": ""
},
"inputFormat": {
"type": "json"
}
},
"tuningConfig": {
"type": "index_parallel",
"partitionsSpec": {
"type": "hashed",
"numShards": 1,
"partitionDimensions": [
"dim"
]
},
"forceGuaranteedRollup": true,
"maxNumConcurrentSubTasks": 2,
"splitHintSpec" : {
"type" : "maxSize",
"maxSplitSize" : 1
}
},
"dataSchema": {
"dataSource": "int_rollup_issue",
"granularitySpec": {
"type": "uniform",
"queryGranularity": "HOUR",
"rollup": true,
"intervals": [
"2020-05-15/2020-05-16"
],
"segmentGranularity": "DAY"
},
"timestampSpec": {
"column": "time",
"format": "posix"
},
"dimensionsSpec": {
"dimensions": [
{
"type": "long",
"name": "dim"
}
]
},
"metricsSpec": [
{
"name": "count",
"type": "count"
},
{
"name": "sum_value",
"type": "longSum",
"fieldName": "value"
}
]
}
}
}
```

Reproduce the native-ingestion case with the two JSON files and the provided ingestion spec, confirming that four rows appear instead of three. Read RollupFactsHolder and RowCombiningTimeAndDimsIterator to trace how integer nulls and default values affect ordering and merging. Done means the reproduced input produces three rows with identical rows combined correctly.

Written by the indexing model from the issue text.