vendor: update leveldb

[rjl493456442]

This PR includes a leveldb fix for long time compaction.

The upstream PR is here. Open this PR for experiment right now.

[karalabe]

Charts: https://app.datadoghq.com/dashboard/u6g-jmw-28b/experimental-64-vs-master-129-2?to_ts=1551086700000&tpl_var_exp=mon07.ethdevops.io&live=true&is_auto=false&tpl_var_master=mon06.ethdevops.io&tile_size=m&from_ts=1551083100000&page=0

Logs: https://papertrailapp.com/groups/11020232/events

[rjl493456442]

A few benchmarking updates:

The experiment branch with fix now is about 100k blocks ahead. And currently no Database compacting, degraded performance log printed at all. While the master branch suffers to write pause about 37min per hour.

Currently, the database size of the experiment branch is 518G, sstable file number is 272317.

block difference diagram: the block difference is gradually increasing.

Data write speed, the below one is master branch. It is obvious that write speed becomes very low at regular intervals. It means disk io is almost idle while compaction is still undergoing

Master branch execution time vs state write time: It is obvious that the duration of data writing takes up almost the entire time required for a block import at regular intervals.

Experiment branch execution time vs state write time: State write time is very stable while execution time is increasing gradually

[karalabe]

LGTM!!!

[karalabe]

I'll merge this PR, more time on master to notice any issues. The benchmarks we can keep running to see how import times evolve over the next few days. Curious how much time a full archive sync will take.