Turn FutureState into an inline type and allocate it using std.ptr.resize

[yorickpeterse]

The FutureState type in std.sync is a heap type, meaning that it also includes a header. The way it's used is such that we never make use of that header, resulting in 16 bytes of wasted space per FutureState.

Now that we have support for inline types we can optimize this as follows:

• We make FutureState an inline type
• We allocate it using std.ptr.resize
• Instead of a UInt64 that we cast to an owned value and then forget about, we use a Pointer[FutureState] and access the fields through that pointer

This will probably require some manual fiddling with pointer offsets for e.g. access to the atomic state

The tricky thing/downside is that we'd depend on FutureState having a specific layout. While this is currently true, it's not something that we guarantee. We can't use an extern type as these can't be generic.