[Query] Share fragment exchange budgets and consume ready Collect inputs

[ColinLeeo]

Description

Exchange memory is divided among individual handles, and Collect consumes children in sequence. With many inputs, this can strand memory in per-handle quotas and leave ready inputs waiting behind a blocked child. This change shares each fragment instance's exchange budget by direction and lets Collect consume a ready input while preserving query correctness after unordered fan-in.

Shared exchange budgets

• Split the fragment exchange budget F into fixed INPUT = F / 2 and OUTPUT = F - INPUT budgets shared by all pipelines in that fragment.
• Charge remote sources to the receiving fragment's INPUT budget and sinks/local shared queues to the producing fragment's OUTPUT budget. Local queues are charged once, independent of creation order.
• Maintain global, fragment-side, and plan-node accounting across reservation, rollback, cancellation, and release. Retry a waiting reservation after publishing it so a concurrent release cannot miss the request; synchronize completion with release/cancellation to prevent duplicate charging.

Ready-input Collect and plan properties

• Collect and MappingCollect observe every live direct child, reuse pending-future listeners, and keep the selected child stable during output backpressure. Operator methods run on the driver thread; completion callbacks only report readiness or failure.
• Refresh readiness after null/empty results, propagate child failures and cancellation, close all inputs, and account for the retained memory of all live children. MappingCollect uses the actual selected child's mapping.
• Clear invalid pre-grouping above unordered Collect/Union so aggregation uses HashAggregation. Preserve valid streaming aggregation in ordered branches, restore partition sorting when required, and prevent sort elimination across unordered fan-in.
• Derive TopK input-order flags from actual physical inputs and preserve runtime filter identities. MergeSort/Join are covered by dependency validation only; their consumption algorithms and progress mechanisms are outside this change. Any-ready consumption for TableTopK and CountMerge is also outside this change.

Design and detailed validation commands: exchange fan-in design.

Validation

• 118 targeted unit tests across 14 classes passed, with no failures or skipped tests. Coverage includes a deterministic free-before-enqueue race, shared-budget contention and cleanup, Collect with 100/1000 children, concurrent readiness, failure/close handling, and grouping/ordering/runtime-filter properties.
• Full-reactor English compilation/tests and Chinese test-compile -DskipTests -P with-zh-locale passed.
• Rebuilt and installed the branch's node artifacts, then ran all 7 IoTDBCollectFanInIT tests in both TableSimpleIT and TableClusterIT (three DataNodes). Both runs passed. SQL coverage includes aggregation/DISTINCT, Union, nested windows, Fill, TopK, tree/table exchange paths, and join dependencies.
• Spotless, Checkstyle, license checks, and git diff --check passed. Maven validation used offline mode with rebuilt local artifacts.

---

This PR has:

• been self-reviewed, including concurrent reservation, release, cancellation, and readiness paths.
• added documentation for modified behaviors.
• added comments explaining non-obvious concurrency and ownership decisions.
• added or updated unit tests.
• added integration tests.
• been tested in a test IoTDB cluster.

Key changed/added classes

MemoryPool, MemorySide, LocalExecutionPlanContext, exchange handles and SharedTsBlockQueue; AnyChildBlocked, CollectOperator, MappingCollectOperator; TableDistributedPlanGenerator, ParallelizeGrouping, and SortElimination.