Improve type definitions for decode

[s1na]

Thinking abstractly about decode I'd expect a function signature such as:

export function decode(input: Buffer): NestedBufferList

where NestedBufferList could be any nested composition of buffers and arrays such as Buffer, or Buffer[], or [Buffer, Buffer[]], etc. (haven't tried expressing this in TS).

Currently the decode method has the following definitions:

rlp/src/index.ts

Lines 60 to 63 in f06a96f

	export function decode(input: Buffer, stream?: boolean): Buffer
	export function decode(input: Buffer[], stream?: boolean): Buffer[]
	export function decode(input: Input, stream?: boolean): Buffer[] | Buffer | Decoded
	export function decode(input: Input, stream: boolean = false): Buffer[] | Buffer | Decoded {

1. First definition takes Buffer and returns a Buffer. Not sure why, it could very well be that the decoding is a NestedBufferList. This was also the problem stated in Encoding and Decoding RLP lists when part of a buffer #77
2. Second def. takes a Buffer[]. Is this to support decoding multiple serialized items simultaneously? the code doesn't seem to do that. I suggest we remove this
3. The last two take Input as argument. I imagine this is for backwards-compatibility. I suggest we tighten the API and only accept a Buffer

On the stream flag: we could keep it, or add a new function decodeStreaming.

[wemeetagain]

An additional wrinkle, the readme says :
rlp.decode(encoded, [skipRemainderCheck=false]) - Decodes an RLP encoded Buffer, Array or String and returns a Buffer or an Array of Buffers. If skipRemainderCheck is enabled, rlp will just decode the first rlp sequence in the buffer. By default, it would throw an error if there are more bytes in Buffer than used by rlp sequence.

It looks like the skipRemainderCheck logic is gone.

EDIT:
the skipRemainderCheck param isn't needed anymore because the stream param does the same thing and more.