Add missing asynchronous utility functions

[h2non]

Ideally, porting the following functions to the asynchronous world:

• itertools.chain
• itertools.chain.from_iterable
• functools.partial_method
• itertools.takewhile
• itertools.compress
• itertools.startmap
• take
• itertools.tee

[h2non]

See: http://fitzgen.github.io/wu.js/

[JamesMcGuigan]

I'd like to suggest functions similar to bluebird.js::props(), and pydash::map_values() for working with dictionaries rather than lists.

This allows computing a series of async functions whilst retaining the original mapping between input and output values, and is also a convenient way to load multiple named async variables before the main synchronous code block.

output = await paco.map_dict({
   "A.txt": async_read_file("A.txt"),
   "B.txt": async_read_file("B.txt")
})
assert output == { "A.txt": "contents A.txt", "A.txt": "contents B.txt" }

output = await paco.map_dict({
   "A.txt": "A.txt",
   "B.txt": "B.txt"
}, async_read_file)
assert output == { "A.txt": "contents A.txt", "B.txt": "contents B.txt" }

output = await paco.map_dict(["A.txt", "B.txt"], async_read_file)
assert output == { "A.txt": "contents A.txt", "B.txt": "contents B.txt" }

[h2non]

Oh, yeah, that's relatively common if you came from a JS background, but not sure if Python devs are going to appreciate this. At the end I feel it's pretty much logic sugar for convenience and laziness.

If any, I would only bet supporting interface and behavior of example 1 and 2, but no more. I prefer a concrete, predictable interface instead of a too overloaded interface with too many vectors.

This can be easily implemented in multiple ways, such as using the async reducer.
PR are welcome.