[core] Support nested-key-null-strategy in FieldNestedUpdateAgg operator

[PyRSA]

Purpose

[Feature] Add fields.{fieldName}.nested-key-null-strategy configuration for nested_update function
[Feature] Support nested-key null handling strategies in FieldNestedUpdateAgg operator to control behavior when nested-key does not satisfy primary key semantics

Expected behavior when nested-key contains null values or does not satisfy primary key semantics:

In the following examples, nested-key is defined as k0,k1.

case	Input A	Input B	Current Behavior	Expected Behavior
Default behavior (no config)	[(0, null, "A", 1)]	[(0, 1, "B", 2)]	Merge	Merge
MERGE strategy	[(0, null, "A", 1)]	[(0, 1, "B", 2)]	Merge	Merge
IGNORE strategy	[(0, null, "A", 1)]	[(0, 1, "B", 2)]	Merge ❌	Ignore invalid nested-key row
ERROR strategy	[(0, null, "A", 1)]	[(0, 1, "B", 2)]	Merge ❌	Throw exception

---

Behavior Definition

• merge
  Merge rows even if nested-key does not satisfy primary key semantics.
  This is equivalent to not configuring nested-key-null-strategy.

• ignore
  Ignore rows whose nested-key does not satisfy primary key semantics.

• error
  Throw an exception when nested-key does not satisfy primary key semantics.

---

Tests

• org.apache.paimon.mergetree.compact.aggregate.FieldAggregatorTest#testFieldNestedUpdateAggWithNestedKeyNullStrategyArgumentCheck
• org.apache.paimon.mergetree.compact.aggregate.FieldAggregatorTest#testFieldNestedUpdateAggWithoutNestedKeyNullStrategy
• org.apache.paimon.mergetree.compact.aggregate.FieldAggregatorTest#testFieldNestedUpdateAggWhenNestedKeyNullUseMergeStrategy
• org.apache.paimon.mergetree.compact.aggregate.FieldAggregatorTest#testFieldNestedUpdateAggWhenNestedKeyNullUseIgnoreStrategy
• org.apache.paimon.mergetree.compact.aggregate.FieldAggregatorTest#testFieldNestedUpdateAggWhenNestedKeyNullUseThrowErrorStrategy
• org.apache.paimon.mergetree.compact.aggregate.FieldAggregatorTest#testFieldNestedUpdateAggWithCountLimitWhenNestedKeyNullUseMergeStrategy
• org.apache.paimon.mergetree.compact.aggregate.FieldAggregatorTest#testFieldNestedUpdateAggWithCountLimitWhenNestedKeyNullUseIgnoreStrategy
• org.apache.paimon.mergetree.compact.aggregate.FieldAggregatorTest#testFieldNestedUpdateAggWithCountLimitWhenNestedKeyNullUseThrowErrorStrategy
• org.apache.paimon.mergetree.compact.aggregate.FieldAggregatorTest#testFieldNestedUpdateAggRetractAppliesNestedKeyNullStrategyToAccumulator
• org.apache.paimon.mergetree.compact.aggregate.FieldAggregatorTest#testFieldNestedUpdateAggRetractAppliesNestedKeyNullStrategyToRetractInput

---

API and Format

No Changes.

---

Documentation

docs/docs/primary-key-table/merge-engine/aggregation.mdx

[JingsongLi]

Why not create a default value?

[PyRSA]

Thanks for the suggestion! I originally avoided a default value because I wanted to distinguish between an unspecified option and an explicitly configured MERGE, so that I could validate that nested-key-null-strategy is only configured when nested-key is present. With a default value, these two cases become indistinguishable. The default behavior is still preserved by falling back to MERGE in FieldNestedUpdateAgg.
I'm happy to adjust this if you think consistency with other options is more important.

[PyRSA]

Why not create a default value?

Thanks for the suggestion! I think using a default value also makes sense.

The only issue is that once MERGE becomes the default, FieldNestedUpdateAgg can no longer distinguish between an unspecified option and an explicitly configured MERGE in its constructor, so the dependency validation cannot be performed there anymore.

Instead, I'm thinking of moving the validation to FieldNestedUpdateAggFactory.create(), where the dependency can be validated before creating the aggregator by checking whether nested-key-null-strategy was explicitly configured. This still provides fail-fast validation during table creation and avoids allowing table options that have no effect.

If this approach sounds reasonable, I'll update the implementation accordingly.

boolean strategyConfigured =
        options.toConfiguration()
                .containsKey(...);

checkArgument(
        !strategyConfigured || !nestedKey.isEmpty(),
        ...);

Something along these lines.