Fix SessionDataSet: row count, async dispose, and null measurement issues (#53, #54, #55)

[CritasWang]

• Unexpected behavior of the CurrentBatchRowCount() method #53 Fix CurrentBatchRowCount() always returning 0 by eagerly constructing the first TsBlock in RpcDataSet constructor when initial data is available 修复 CurrentBatchRowCount() 始终返回 0 的问题，在构造函数中预先反序列化首个 TsBlock

• Why doesn’t SessionDataSet implement the IAsyncDisposable interface? #54 Add IAsyncDisposable to SessionDataSet and RpcDataSet, providing DisposeAsync() that properly awaits Close() to avoid sync-over-async deadlocks 为 SessionDataSet 和 RpcDataSet 添加 IAsyncDisposable 接口，支持 await using 语法

• Unexpected behavior of the Next() method in SessionDataSet #55 Fix GetRow() including null-valued columns in RowRecord by using IsNull() check before calling type-specific getters, instead of relying on value type null checks which always pass for int/bool/float/etc. 修复 GetRow() 中值类型默认值绕过 null 检查导致空值列被错误包含的问题

[HTHou]

Thanks for the fixes. The main direction looks good: eager TsBlock construction addresses CurrentBatchRowCount(), and the explicit null check in GetRow() fixes the value-type default-value issue.

One thing I think should be fixed before merge: SessionDataSet.Close() still never sets _isClosed = true. After the first successful close it also sets _client = null, so a second call to Close() can enter the same branch again and dereference _client.SessionId. This becomes more visible now that both Dispose() and DisposeAsync() route through Close(). Could we make Close() idempotent, for example by marking _isClosed = true in the close path before returning the client and avoiding re-adding a null/already-returned client?

It would also be great to add regression tests for:

• CurrentBatchRowCount() returning the first TsBlock size before reading the first row.
• GetRow() excluding null-valued measurements for value types like BOOLEAN, INT32, and DOUBLE.

[CritasWang]

Thanks for the fixes. The main direction looks good: eager TsBlock construction addresses CurrentBatchRowCount(), and the explicit null check in GetRow() fixes the value-type default-value issue.

One thing I think should be fixed before merge: SessionDataSet.Close() still never sets _isClosed = true. After the first successful close it also sets _client = null, so a second call to Close() can enter the same branch again and dereference _client.SessionId. This becomes more visible now that both Dispose() and DisposeAsync() route through Close(). Could we make Close() idempotent, for example by marking _isClosed = true in the close path before returning the client and avoiding re-adding a null/already-returned client?

It would also be great to add regression tests for:

• CurrentBatchRowCount() returning the first TsBlock size before reading the first row.
• GetRow() excluding null-valued measurements for value types like BOOLEAN, INT32, and DOUBLE.

fixed

[HTHou]

Thanks, the requested idempotency fix and regression tests are in place. CI is green and the changes look good to me.