perf: optimize checksum calculation with IncrementalHash

[olavloite]

Use IncrementalHash for calculating the checksum of a transaction, rather than creating a new SHA256 instance for every row.

[gemini-code-assist]

Code Review

This pull request refactors SpannerDataReaderWithChecksum to use IncrementalHash (SHA256) for running checksum calculations instead of re-calculating it from scratch on every row, which improves performance and reduces allocations. It also caches the final checksum upon disposal to support transaction retries. The reviewer suggests optimizing the CodedOutputStream instantiation in AppendRowToHash by specifying a smaller buffer size (e.g., 256 bytes) to further reduce per-row memory allocations and mitigate GC pressure.