Turns the mirrored module into an allocator

[gnzlbg]

The allocator is currently trivial. It never frees any memory. Instead it caches the freed allocations into a HashMap<size, Vec<*mut ptr>>. When a new allocation is requested, it tries to reuse a previously-freed allocation from the cache.

A slice_deque::release() function is provided that frees all the currently unused memory.

[gnzlbg]

@bill-myers what do you think of this? It basically caches the smallest allocation in a vector, and the rest in a HashMap, and does so in thread-local caches.

[Boscop]

@gnzlbg Wouldn't it be faster to use fnv as the hasher for the HashMap? I'm no expert but fnv is usually recommended for small/integer keys.