This bug sweeps up some of the cases remaining after bug 914814 that don't use ImmPtr for embedding pointers in generated code.

Rather than switching to ImmPtr (which will require special treatment to in bug 900669), this patch removes loadStatic(Double|Float), replacing them with loadConstant(Double|Float). This should generate equivalent code with the only difference being slightly more memory to store the double/float (which seems negligible).

In one other case, we load from static doubles for the operand to an add. This patch adds an addConstant(Double|Float) to achieve the same effect.

This patch removes the void*-taking overload of Operand (requiring instead Operand(AbsoluteAddress(ptr)). Several of these cases involve generating loads from an address that is patched in later. Since this is fine for serialization (the patching will also happen after deserialization), I introduced a new PatchedAbsoluteAddress. The key distinction is that, while AbsoluteAddress/ImmPtr will JS_ASSERT(!IsCompilingAsmJS()), PatchedAbsoluteAddress won't.

This patch (I think the last one) switches j/addPendingJump to use ImmPtr.

https://hg.mozilla.org/integration/mozilla-inbound/rev/fa389e608d2a https://hg.mozilla.org/integration/mozilla-inbound/rev/53edc4194eb3 https://hg.mozilla.org/integration/mozilla-inbound/rev/e482e83c5e88

https://hg.mozilla.org/mozilla-central/rev/fa389e608d2a https://hg.mozilla.org/mozilla-central/rev/53edc4194eb3 https://hg.mozilla.org/mozilla-central/rev/e482e83c5e88