[c++] Add bounded RecordBatch log reader APIs

[naivedogger]

Purpose

Linked issue: close #3987

Expose bounded RecordBatchLogReader APIs through the C++ binding.

API example

fluss::RecordBatchLogReader reader;
check("create_bounded_reader",
      table.NewScan().CreateRecordBatchLogReader(
          admin,
          assigned_buckets,
          fluss::TimestampRange{
              starting_timestamp_ms,
              stopping_timestamp_ms,
          },
          reader));

while (true) {
    fluss::RecordBatchReadResult result;
    check("next_batch", reader.NextBatch(1000, result));

    if (result.status == fluss::BoundedReadStatus::TimedOut) {
        continue;
    }
    if (result.status == fluss::BoundedReadStatus::Finished) {
        break;
    }

    process(result.batch->GetArrowRecordBatch());
}

The API also supports:

• explicit starting and stopping offsets for each bucket;
• starting offsets configured through scanner subscriptions;
• stopping at explicit or latest offsets;
• partitioned and non-partitioned tables.

TimedOut allows callers to check cancellation or deadlines before retrying. Finished means all assigned buckets have reached their stopping offsets.

Tests

• cargo test -p fluss-rs --lib client::table::reader
• cargo test -p fluss-cpp --lib
• cargo clippy -p fluss-rs -p fluss-cpp --lib -- -D warnings
• cargo fmt --check
• CMake C++ build
• Docker-backed C++ integration tests for explicit offsets, latest offsets,
  timestamp ranges, and partitioned tables

API and Format

This PR adds backward-compatible public C++ APIs.

Documentation

Updated the C++ README, API reference, log table guide, and example program.

[naivedogger]

@fresh-borzoni @charlesdong1991 @leekeiabstraction Appreciate a review here when you have time 🙏

[leekeiabstraction]

TY for the PR, added a couple of comments. PTAL.

[leekeiabstraction]

When FFI call fails, we return with out.status still set to TimedOut, this will cause caller following documentation to retry forever on unretriable errors.

[naivedogger]

Fixed. If the FFI call fails, NextBatch() now leaves out as Finished with no batch instead of TimedOut. This prevents callers from mistaking an actual error for a timeout and retrying forever. Callers should still check the returned Result before using out.

[leekeiabstraction]

This doesn't have a timeout arg, what happens if it is an error such as a tablet server outage? Will this hang caller forever?

[naivedogger]

Fixed. CollectAllBatches(timeout_ms, out) now uses timeout_ms as the time limit for the entire collection. If the timeout expires before all stopping offsets are reached, it stops and returns REQUEST_TIME_OUT; any complete batches already collected remain in out. Only Ok() means the result is complete. The documentation, examples, and tests have been updated accordingly.

[naivedogger]

Thanks @leekeiabstraction for the thorough review! Both comments have been addressed and the updates have been pushed. PTAL when you have a chance.