Support flexible numL0Partitions

[ZexiLiu]

Currently numL0Partitions is fixed and once created a DB with a certain number, we cannot change the number because TableMgr will get the number of L0 partition from table_manifest.

To support flexible numL0Partitions, I think of two solutions:

1. Upon each TableMgr::init(): Manifest.numL0Partitions means numL0Paritions get from table manifest

if (Manifest.numL0Partitions== DBConfig.numL0Partition) {
    skip below logic
}

// Manifest.numL0Partitions != DBConfig.numL0Partition
if (Manifest.numL0Partitions > DBConfig.numL0Partition) {
    compact the first (Manifest.numL0Partitions - DBConfig.numL0Partition) L0 tables to L1 
} else {
    // Manifest.numL0Partitions < DBConfig.numL0Partition
    create (DBConfig.numL0Partition - Manifest.numL0Partitions) empty L0 tables
}

run DB as before like the numL0Partition is fixed and defined by DBConfig. 

2. Let compactor handle the adjustment to the difference.

I think solution 1 is much easier to implement than 2. And there is no need to support DBConfig.numL0Partition as hot swapable config.

Please kindly let me know your opinion on this. Thanks.

[ZexiLiu]

#180

PR for solution 1.

[greensky00]

The PR is merged, but one potential issue remains. If multiple threads invoke DB::open, only one thread should adjust L0 partitions, while the others should be blocked. We may need to use compare_exchange_strong and mutex to support it.