Optimise the heap data structure.

I'm not familiar enough with this projects finer details and intent behind `src/native/heap.rs`, so my assumptions may be a bit off. I have, however, recently been doing some work that looks to me like it relates quite closely with this implementation: A heap backed priority queue with removal of arbitrary entry.

To start, [it's commented](https://github.com/async-rs/futures-timer/blob/30e562cdde45b8a6ee93d22094df8664ff8dfdfb/src/native/heap.rs#L8) that things aren't optimised. I would like to discuss how I might bring in some optimizations I have in mind.

* I can see that the heaps sink/swim functions are manually defined. If you compare it with `std-lib` implementation, it has an interesting use of a data struct with custom `Drop`s, which seems to improve things through minimizing movement of values in memory. I'd like to try and tap into the std-lib implementation, otherwise port it in and see if it improves things.
* The heap appears to be an array of references into a linked list. The linked list itself is likely to have cache-unfriendly memory layout. The change I have in mind allows for cache-friendly arrays.
* It looks like entries are identified by index. A change I have in mind should allow indexing to be kept, and also allow for lookup by unique id based on hash-map lookup: O(1)~

I would like to discuss these things in more detail before I get started, so I keep within the design intent of the system.

Start by reading src/native/heap.rs and comparing its sink/swim implementation with the standard-library heap implementation mentioned in the issue. Review the proposed cache-friendly storage and unique-id lookup ideas with maintainers first; the issue does not specify tests or a concrete completion criterion.

Written by the indexing model from the issue text.