Remove the stop watch usage during compactions.

[psudheer21]

While investigating a performance bottleneck we noticed that almost all the worker threads were performing clock_gettime() with the following stack trace:

 @     0x7fc84f95411e  (unknown)
    @     0x7ffff07c67c1  ([vdso]+0xffffffffff7007c1)
    @     0x7fc84fa13fe4  __GI___clock_gettime
    @     0x7fc85979e9f4  rocksdb::(anonymous namespace)::PosixEnv::NowNanos()
    @     0x7fc859632eeb  rocksdb::StopWatchNano::StopWatchNano()
    @     0x7fc85964a408  rocksdb::CompactionIterator::NextFromInput()
    @     0x7fc85964ac68  rocksdb::CompactionIterator::Next()
    @     0x7fc859650083  rocksdb::CompactionJob::ProcessKeyValueCompaction()
    @     0x7fc8596516cd  rocksdb::CompactionJob::Run()
    @     0x7fc85968e4ef  rocksdb::DBImpl::BackgroundCompaction()
    @     0x7fc8596900fd  rocksdb::DBImpl::BackgroundCallCompaction()
    @     0x7fc8596a1be6  rocksdb::DBImpl::CompactionTask::DoRun()
    @     0x7fc859696607  rocksdb::DBImpl::ThreadPoolTask::Run()
    @     0x7fc855a690a6  yb::(anonymous namespace)::PriorityThreadPoolWorker::Run()
    @     0x7fc855a9430d  yb::Thread::SuperviseThread()
    @     0x7fc8502ca692  start_thread

This usage of the stop watch is done just for stats and is something we can remove.