move TableCache::EraseHandle outside of db mutex

Summary:
Post-compaction work holds onto db mutex for the longest time (found by tracing lock acquires/releases with LTTng and correlating timestamps with our info log). Further experimentation showed `TableCache::EraseHandle` is responsible for ~86% of time mutex is held. We can just release the handle outside the db mutex.
Closes https://github.com/facebook/rocksdb/pull/2654

Differential Revision: D5507126

Pulled By: ajkr

fbshipit-source-id: 703c01ddf2aea16bc0f9e33c08935d78aa6b781d

db/db_impl_files.cc

@@ -368,6 +368,9 @@ void DBImpl::PurgeObsoleteFiles(const JobContext& state, bool schedule_only) {
     candidate_files.emplace_back(
         MakeTableFileName(kDumbDbName, file->fd.GetNumber()),
         file->fd.GetPathId());
+    if (file->table_reader_handle) {
+      table_cache_->Release(file->table_reader_handle);
+    }
     delete file;
   }
 

db/version_set.cc

@@ -328,10 +328,6 @@ Version::~Version() {
       assert(f->refs > 0);
       f->refs--;
       if (f->refs <= 0) {
-        if (f->table_reader_handle) {
-          cfd_->table_cache()->EraseHandle(f->fd, f->table_reader_handle);
-          f->table_reader_handle = nullptr;
-        }
         vset_->obsolete_files_.push_back(f);
       }
     }