Fix the OOM in log_server

[kvinwang]

As @shelvenzhou reported, the sidevm of the contract log_server panics due to OOM.

See #1325 for the misleading of the error message.

I've checked it with random test logs and didn't see memory leak in the sidevm program.

The size of logs was limited with a soft bound of 8MB. However, we under-estimated the current log size with a rough formula.
Counting in the overhead, the actual memory cost was far more than 8MB. And a large tmp memory blob is required when a query requests all log records. This could make the guest run out of memory.

This PR makes 2 improvements:

• Using the metric reported from the memory allocator as the records size bounds. This should be more accurate than the estimated one before.
• Boxing each single log record. This could reduce the overhead of the VecDeque container.

[h4x3rotab]

LGTM