row_cache: Evict partition snapshots

If snapshots are not evicted, they may pin unbouned amount of memory
for a long time in cache, which may lead to OOM. Evict snapshots
together with the entry.

Fixes #2775.
Fixes #2730.

row_cache.cc

@@ -854,7 +854,7 @@ future<> row_cache::update_invalidating(external_updater eu, memtable& m) {
             // This invalidates all row ranges and the static row, leaving only the partition tombstone continuous,
             // which has to always be continuous.
             cache_entry& e = *cache_i;
-            e.partition() = partition_entry(mutation_partition::make_incomplete(*e.schema(), mem_e.partition().partition_tombstone()));
+            e.partition().evict(); // FIXME: evict gradually
         } else {
             _tracker.clear_continuity(*cache_i);
         }
@@ -966,6 +966,10 @@ cache_entry::cache_entry(cache_entry&& o) noexcept
     }
 }
 
+cache_entry::~cache_entry() {
+    _pe.evict();
+}
+
 void row_cache::set_schema(schema_ptr new_schema) noexcept {
     _schema = std::move(new_schema);
 }

row_cache.hh

@@ -124,6 +124,7 @@ public:
     }
 
     cache_entry(cache_entry&&) noexcept;
+    ~cache_entry();
 
     bool is_evictable() { return _lru_link.is_linked(); }
     const dht::decorated_key& key() const { return _key; }