feat(partial-update): support field aggregation reads

[shyjsarah]

Motivation

Paimon Rust can now read merge-engine=partial-update tables with sequence groups after #584, but it still rejects existing Java/Flink-created tables that configure:

• fields.<field>.aggregate-function
• fields.default-aggregate-function

This prevents Rust and DataFusion readers from opening partial-update tables that aggregate protected fields within sequence groups. Treating these tables as basic latest-non-null partial update would be incorrect, especially when an older sequence-group row must be combined in reverse order.

I searched the existing issues and did not find one covering partial-update field aggregation reads. This is a follow-up to #583 / #584.

Proposed scope

Add read-side field aggregation support for partial-update tables:

• parse and validate per-field and default aggregate-function options;
• reuse the aggregate functions already implemented in Paimon Rust;
• support last_non_null_value without a sequence group, matching Java validation;
• require other configured aggregate functions to protect fields in a sequence group;
• apply normal aggregation when an incoming group sequence is newer or equal;
• apply Java-compatible reversed aggregation when an incoming group sequence is older;
• preserve sequence-group projection dependency widening;
• fail closed with a clear error for unknown or type-incompatible aggregate functions;
• add merge-level and end-to-end projected-read tests.

Out of scope

To keep the contribution reviewable:

• adding Java-only aggregate-function families that are not currently implemented in Rust;
• DELETE / UPDATE_BEFORE retraction and remove-record semantics;
• Rust writer and CREATE/ALTER support for partial-update aggregation.

Willingness to contribute

I'm willing to submit a PR for this scoped read-side support.