Literal object creation rework

[andreabergia]

This PR changes a bit the interpreter bytecode when creating a literal object - rhino now generates and uses a slightly more compact bytecode. Furthermore, there is a preliminary change for implementing super discussed below.

Before this change, this JS fragment:

var o = {
  a: 1,
  [b]: 42,
}

would generate the following interpreter bytecode:

 [5] REG_IND_C0
 [6] LITERAL_KEYS [a, 136] true
 [8] REG_IND_C2
 [9] LITERAL_NEW
 [10] ONE
 [11] LITERAL_SET
 [12] REG_STR_C1 "b"
 [13] NAME
 [14] REG_IND_C1
 [15] LITERAL_KEY_SET
 [16] SHORTNUMBER 42
 [19] LITERAL_SET
 [20] OBJECTLIT

After this PR, it will generate the following:

 [5] REG_IND_C0
 [6] LITERAL_NEW_OBJECT [a, 142] true
 [8] ONE
 [9] LITERAL_SET
 [10] REG_STR_C1 "b"
 [11] NAME
 [12] LITERAL_KEY_SET
 [13] SHORTNUMBER 42
 [16] LITERAL_SET
 [17] OBJECTLIT

The changes are:

• LITERAL_KEY_SET uses the same indexing algorithm as LITERAL_SET, therefore does not require the REG_IND_xxx instruction before it
• renamed old LITERAL_NEW into LITERAL_NEW_ARRAY
• unified LITERAL_KEYS and the old behavior of LITERAL_NEW for objects into a new opcode called LITERAL_NEW_OBJECT
• modified LITERAL_NEW_OBJECTS so that it puts the object on the stack and then calls ScriptRuntime.fillObjectLiteral, instead of calling ScriptRuntime.newObjectLiteral. So, previously the stack would contain [keys, flags, values]; now it contains [object, keys, flags, values]. This is a preliminary step for implementing super (see Implement super #1735).

Implemented with @0xe