perf(spanner): replace setTimeout backpressure with event-driven resumption

[olavloite]

Replace the setTimeout backpressure polling loop in PartialResultStream with native Node.js stream flow control:

• Hold the transform completion callback when downstream push() returns false, and invoke it upon stream drain via _read() to resume upstream ingestion.
• Remove _tryResume, _numPushFailed, and the artificial 'Stream is still not ready to receive data' timeout failures.
• Guard 'paused' event emission in _addChunk so it only emits once per backpressure transition instead of on every subsequent value.
• Mark maxResumeRetries as deprecated on streaming RowOptions, clarifying that it is only used by non-streaming Snapshot.run() for RPC retry limits.
• Replace setTimeout with setImmediate in test/spanner.ts slow-writer test, eliminating a 5-second sleep in compliance with the zero-sleep test policy.
• Add comprehensive backpressure unit tests covering single-chunk direct decoding, multi-chunk pause/resume cycles, cross-chunk value stitching, stream destroy/error propagation, and full-pipeline flow control.

[gemini-code-assist]

Code Review

This pull request replaces the timeout-based retry mechanism for stream backpressure with automatic, event-driven stream flow control in PartialResultStream, deprecating the maxResumeRetries option for streaming queries. Feedback on the changes suggests wrapping super._read(size) in an else block inside the _read method to avoid potential redundant calls to the parent class's read implementation when resuming from a paused state.

[olavloite]

/gemini review

[gemini-code-assist]

Code Review

This pull request refactors backpressure handling in PartialResultStream to use native Node.js stream flow control instead of a timer-based retry mechanism. The maxResumeRetries option is deprecated for streaming queries, and the stream now pauses and resumes automatically using _read() and a cached callback when downstream buffers are full. Comprehensive unit tests have been added to verify this event-driven backpressure behavior, and existing tests have been updated to reflect that slow writers no longer trigger failures. There are no review comments to address, and the implementation looks solid.