[rlp] consider adding a way to allocate less buffer space

[dvdplm]

Currently rlp lists allocate a new RlpStream which allocates a Vec<u8> with a capacity of 1024. For many calls the size is much smaller than this and can often be calculated easily at the call site. Consider adding some way to ask for a bespoke buffer size something like new_list_with_capacity(4, 86) (4 items, initial capacity of 86 bytes). Some calls likely need plenty more than 1024.

e.g. https://github.com/paritytech/parity-ethereum/blob/master/ethcore/sync/src/chain/requester.rs#L82 (from sync) or https://github.com/paritytech/parity-ethereum/blob/master/util/patricia-trie-ethereum/src/rlp_node_codec.rs#L155 (from patricia-trie).

Bonus points for collecting stats about RLP sizes actually used, maybe using historian?