A GL Query cache of 1024 items is enough (was 4096).

Since each QuerySyncManager has its own cache, even a moderate
cache size expands to several hundred KB of memory usage. This makes
it slightly smaller to save some of that memory. It is believed
to be unlikely that the cache has to be big anyway.

In ad-hoc testing I've only seen 20 queries in total (unclear how
many of them were in parallel).

(This is alt #1 in bug 485536)

BUG=485536
R=reveman@chromium.org

Review URL: https://codereview.chromium.org/1138463003

Cr-Commit-Position: refs/heads/master@{#329383}

gpu/command_buffer/client/query_tracker.h

@@ -27,7 +27,7 @@ class GLES2Implementation;
 // Manages buckets of QuerySync instances in mapped memory.
 class GLES2_IMPL_EXPORT QuerySyncManager {
  public:
-  static const size_t kSyncsPerBucket = 4096;
+  static const size_t kSyncsPerBucket = 1024;
 
   struct Bucket {
     explicit Bucket(QuerySync* sync_mem)