[Rust] Enforce 10 MiB client-side payload limit on ingest calls

[elenagaljak-db]

What changes are proposed in this pull request?

Enforce the Zerobus server's 10 MiB payload limit client-side on ingest_record_offset and ingest_records_offset. Without this, an oversized payload travels through the landing zone and gRPC layer before being rejected by the server. With this change, the SDK returns ZerobusError::InvalidArgument immediately, before any network I/O.

The limit applies to the total encoded byte size of a single call (sum of all record bytes in the batch). It does not apply to Arrow Flight (ingest_batch / ingest_ipc_batch), which has a separate code path.

Perf impact: measured with Criterion: the check costs ~3 ns for a single record and ~20 ns for a 100-record batch (reading .len() from a SmallVec, no allocation). Well within the noise of the mutex + gRPC overhead.

Changes

• sdk/src/lib.rs — module-level MAX_INGEST_PAYLOAD_BYTES constant (10 MiB) with a note that it mirrors the server limit; check at the top of ingest_internal_v2
• sdk/src/record_types.rs — EncodedBatch::total_byte_size() (pub) + unit tests
• tests/src/rust_tests.rs — integration tests for single-record and batch rejection
• README.md — new "Payload Size Limit" subsection under Error Handling
• NEXT_CHANGELOG.md — entry under New Features and Improvements

How is this tested?

cargo test --workspace