Bug fix: JSON document size reported by 'json.debug memory'

[joehu21]

Fix 2 bugs related to json document size reported by json.debug memory:

1. In command JSON.NUMINCRBY and JSON.NUMMULTBY, the doc size meta data is not updated. This may lead to the doc size meta data becomes negative. For example, when JSON.NUMINCRBY updates an integer value to a double value, the memory size of the key is increased (because double is stored as string). However, the doc size meta data is not updated due to the bug. Later, when JSON.SET is called to set the value to an integer (like 0) or a double that occupies less memory, the key's memory size shrinks and hence a few bytes are subtracted from the doc size meta data. If the process repeats, eventually the doc size meta data becomes negative and is wrapped around as an insanely large positive integer.
2. The value returned from json.debug memory <key> doesn't match the value returned from json.debug memory <key> ..

Code changes:

• RapidJSON: Added GenericValue::ComputeMallocedSize() that computes malloc'ed size of a JValue
• Fix JSON.NUMINCRBY and JSON.NUMMULTBY: update doc size meta data

Test:
Added two python tests:

• test_doc_mem_size_with_numincrby: This test reproduces the bug of "doc size meta data becomes negative".
• test_verify_doc_mem_size: This test verifies the returned from json.debug memory <key> doesn't match the value returned from json.debug memory <key> ..