[python] Add adaptive scalar index reads for LIMIT queries

[XiaoHongbo-Hope]

Purpose

Avoid planning all uncovered row-ID ranges for limited scalar-index lookups on data-evolution append tables.

full remains unchanged. This PR adds adaptive, which:

1. resolves and pins one snapshot;
2. evaluates the scalar index and reads indexed ranges first;
3. applies the complete reader-side predicate and filters;
4. skips uncovered-range planning when the filtered result reaches the limit;
5. otherwise reads uncovered ranges with only the remaining limit.

Reads without a finite limit and static new_scan().plan() calls retain full semantics. The optimization is exposed through ReadBuilder.to_arrow() and multimodal scan(...).to_arrow() / read_blobs().

When native planning is enabled, the small indexed row-range plan can still use pypaimon-rust; fallback ranges stay lazy until they are needed.

Correctness

• Both phases use the same snapshot ID.
• The decision uses rows remaining after the complete reader filter, not index cardinality.
• Indexed and uncovered coverage ranges are disjoint, so the two phases do not duplicate row IDs.
• Unsupported reads conservatively use the existing full path.

Verification

• PyPaimon global-index and multimodal tests: 137 passed
• Java CoreOptionsTest: 17 passed
• Flake8, compileall, Spotless, and git diff --check: passed

Added coverage for index hits, index misses, residual-filter rejection, partial-limit fallback, snapshot pinning, no-limit fallback, and static-plan semantics.