feat(core): add SystemRead for Native metadata reads

[JingsongLi]

Summary

Expose the 11 system tables already supported by PyPaimon through a small Rust core reader:

let mut read = table.new_system_read("files")?;
read.with_projection(&["file_name", "row_count"])?;
read.with_filter(predicate)?;
read.with_limit(100);
let fields = read.read_type();
let batch = read.read().await?;

System metadata is small enough to materialize in one RecordBatch. There are no separate system-table, builder, scan, plan or split types in the core/Native API.

• Move metadata schemas and extraction from DataFusion providers into Rust core. DataFusion and PyO3 adapt SystemRead.
• Support snapshots, schemas, options, manifests, files, buckets, tags, branches, file_key_ranges, table_indexes and consumers. Catalog-dependent partitions retain their existing path.
• Reuse the base table's existing FileIO, REST credentials, branch and captured snapshot. The returned batch is self-contained and preserves its schema when empty.
• Filter before output projection and limit; predicates can reference unprojected columns. Preserve consumer ID point lookup. DataFusion retains residual filtering and does not push a limit ahead of it.
• Expose the same SystemRead methods in Python, with read() returning a PyArrow RecordBatch. PyPaimon stores the full result in its existing in-memory SystemSplit, preserving pickle and offline worker reads.
• Complete Java metadata semantics: persisted latest-schema options, snapshot writer_version, deleteRowCount, composite index fields, local metadata instants and physical-column/schema-evolved statistics. File keys/statistics use Java object strings; partitions use Java CAST strings.

Verification

• Core metadata unit tests: 27 passed.
• DataFusion system-table unit tests: 10 passed.
• DataFusion system-table integration tests: 28 passed.
• Built the companion Python extension with maturin. REST/system-table suites: 165 passed under UTC and 165 passed under America/New_York.
• Covers empty/populated tables, fixed/empty/tag views, branches, PK ranges, vertical writes, schema evolution, consumer pruning, indexes, authorization, filters on unprojected columns, limits, repeated/zero-column projections and captured Python split pickle without REST access.
• Workspace Clippy with Rust 1.98.0 and fulltext/vortex, rustfmt and diff checks passed.
• Independently reviewed the simplified API against the previous implementation: no remaining P1/P2 findings.

Companion: apache/paimon#10604 (Draft). Python CI continues to build Apache paimon-rust main and should merge after this Rust change.

[leaves12138]

LGTM