Preparation for Light client RPC

[rphmeier]

This PR gets us most of the way to RPCs for the current light client, and ports the existing full RPCs over to more generic abstractions instead of MiningBlockChainClient and MinerService.

@tomusdrw This touches a lot of code in the signer implementation (particularly to move it to futures). The tests pass, but let me know if you see anything suspicious. There will also undoubtedly be conflicts with #4477.

Key points:

• Skeleton of eth API
• Generic Dispatcher with FullDispatcher implementation
• OnDemand improvements

Next up:

• Light client transaction queue
• OnDemand service request batching
• LightDispatcher (fetch nonce, historic gas price via OnDemand)

Part of #2235

[rphmeier]

@tomusdrw Grumbles (except for the test stuff) should be addressed now.

[gavofyork]

not very happy about the additional of boilerplate. is there nothing that can be done?

[rphmeier]

It serves a purpose: convert result-using code into future-using code. Two options:

• use the just-written take_weakf! and pay more interleaved "boilerplate" cost with stuff like return future::err(e).boxed() instead of return Err(e)
• write a macro to wrap the function body and compile down to the let inner = || {...} pattern used explicitly here. Still introduces the visual noise + rightwards drift of the status quo.

[tomusdrw]

There is no additional cost with teak_weakf!, in both examples we return one .boxed() future. There is an issue to get rid of those Weak references: #3958 that would also reduce the boilerplate.

[rphmeier]

I understand there's no more runtime cost, I mean you still pay a boilerplate cost in places, just more spread out.

[rphmeier]

Ok, no more let inner pattern via take_weakf and try_bf macros.

[tomusdrw]

Some conflicts to resolve.

[rphmeier]

(resolved locally, but can't build after #4486)

edit: just needed #4507