Specialize Dict's key to ByteArray

[KtorZ]

Looking at Aiken's repositories in the wild, no one is using Dict with
anything else than ByteArray as key. The extra complexity we bring
just for making it generic in the key argument seems not worth it.

Plus, if we know the key is a ByteArray, we can perform some
optimizations on a few functions looking for keys to short-circuit
them instead of traversing a list entirely.

Here's a comparison of the mem and cpu cost before and after the
chance for a few tests (omitted tests have identical cost).

mem	cpu	bench
-10.15%	-5.88%	delete_3
-4.00%	-9.68%	delete_4
-3.80%	-9.39%	delete_5
-31.03%	-42.88%	delete_6
-15.26%	-26.34%	find_4
-11.10%	-18.85%	from_list_2
-8.02%	-16.26%	from_list_3
-12.15%	-25.39%	from_list_4
-19.24%	-36.11%	bench_from_list
-17.79%	-18.91%	bench_from_ascending_list
-4.91%	-10.58%	get_2
-9.49%	-14.63%	get_3
-29.94%	-42.75%	get_4
-23.66%	-37.70%	get_5
-7.94%	-5.03%	has_key_3
-9.97%	-18.50%	has_key_4
-16.46%	-23.61%	insert_1
-16.52%	-26.24%	insert_2
-13.82%	-22.08%	insert_with_1
-18.77%	-29.88%	insert_with_2
-19.12%	-30.52%	insert_with_3
-10.47%	-17.40%	keys_2
4.25%	5.06%	map_1
2.94%	3.31%	map_2
-10.99%	-19.83%	size_3
-7.94%	-16.17%	union_1
-17.74%	-28.17%	union_3
-16.96%	-26.55%	union_4
-15.91%	-25.10%	union_with_1
-10.47%	-17.39%	values_2