Push compute engine value loading for dense singleton ordinals down to tsdb codec.

[martijnvg]

WIP Note: no tests yet, exploring how to implement bulk loading of dense singleton sorted (set) doc values.

This change targets reading field values in bulk mode at codec level when doc values type is sorted doc values or sorted set doc values, there is only one value per document, and the field is dense (all documents have a value).

Relates to #128445

[martijnvg]

The builder is backed by a int[] so we need to do a conversion here, because TSDBDocValuesEncoder is long[] based. But is there a better way? For example can we have builder that accepts long[]? And then in the build() do the conversion to int?

[martijnvg]

I think we need to do this conversation somewhere. Maybe here is good enough.

[martijnvg]

This method is largely the same as the read(...) method except here (decoder.decodeOrdinals(valuesData, currentBlock, bitsPerOrd);) and at the end because the builder is different.

Ideally would like to see less code duplication here.

[martijnvg]

Ordinals are implemented in the codec as numeric doc values. Builder gets created here, since we need to reference a SortedDocValues instance.

[martijnvg]

Depending on whether the field is empty, has a single unique value, sparse or dense we got another implementation here. Only the dense implementation support bulk loading, but that is why this supportsBlockRead() method exists.

[martijnvg]

I've locally benchmarking with the following query: FROM metrics-hostmetricsreceiver.otel-default | STATS count(*) BY host.name | LIMIT 10000. Without this change the total query time is ~260ms and with this change ~150ms.

When running the following profile query:

POST /_query
{
    "profile": true,
    "query": "FROM metrics-hostmetricsreceiver.otel-default | STATS count(*) BY host.name | LIMIT 10000",
    "pragma": {
        "data_partitioning": "shard"
    }
}

Without this change host.name value loading takes:

{
    "operator": "ValuesSourceReaderOperator[fields = [host.name]]",
    "status": {
        "readers_built": {
            "host.name:column_at_a_time:BlockDocValuesReader.SingletonOrdinals": 58
        },
        "values_loaded": 221184000,
        "process_nanos": 1216242904, <-- ~1216 ms
        "pages_received": 45597,
        "pages_emitted": 45597,
        "rows_received": 221184000,
        "rows_emitted": 221184000
    }
}

Wit this change host.name value loading takes:

{
    "operator": "ValuesSourceReaderOperator[fields = [host.name]]",
    "status": {
        "readers_built": {
            "host.name:column_at_a_time:BlockDocValuesReader.SingletonOrdinals": 58
        },
        "values_loaded": 221184000,
        "process_nanos": 387031891, <-- 387 ms
        "pages_received": 45597,
        "pages_emitted": 45597,
        "rows_received": 221184000,
        "rows_emitted": 221184000
    }
}