Object hashes (#104)

* factoring out object_info for general use by several Ray components

* addressing comments

* Replace SHA256 task hash with MD5

Add object hash to object table (always overwrites)

Support for table operations that span multiple asynchronous Redis
commands

Add a new object location in a transaction, using Redis's optimistic
concurrency

Use Redis GETSET instead of transactions and Python frontend code for object hashing

Remove spurious log message

Fix for object_table_add

Revert "Replace SHA256 task hash with MD5"

This reverts commit e599de473c8dad9189ccb0600429534b469b76a2.

Revert to sha256

Test case for illegal puts

Use SETNX to set object hashes

Initialize digest with zeros

Initialize plasma_request with zeros

* Fixes

* replace SHA256 with a faster hash in the object store

* Fix valgrind

* Address Robert's comments

* Check that plasma_compute_object_hash succeeds.

* Don't run test_illegal_put test with valgrind because it causes an intentional crash which causes valgrind to complain.

* Debugging after rebase.

* handling Robert's comments

* Fix bugs after rebase.

* final fixes for Stephanie's PR

* fix

src/common/task.c

@@ -101,13 +101,13 @@ task_id compute_task_id(task_spec *spec) {
   }
   /* Compute a SHA256 hash of the task_spec. */
   SHA256_CTX ctx;
-  BYTE buff[SHA256_BLOCK_SIZE];
+  BYTE buff[DIGEST_SIZE];
   sha256_init(&ctx);
   sha256_update(&ctx, (BYTE *) spec, task_spec_size(spec));
   sha256_final(&ctx, buff);
   /* Create a task ID out of the hash. This will truncate the hash. */
   task_id task_id;
-  CHECK(sizeof(task_id) <= SHA256_BLOCK_SIZE);
+  CHECK(sizeof(task_id) <= DIGEST_SIZE);
   memcpy(&task_id.id, buff, sizeof(task_id.id));
   return task_id;
 }