An account can now approve a producer, and that producer's votes are
increased appropriately. The account can remove approval, and the votes
are decreased appropriately. Additionally, the producer is added/removed
to the account's approved producers list as desired.

TODO: A producer with more votes than the rest is still never rotated in

This new interface allows the chain_controller to request a fresh round
of producers and request the producers' median voted blockchain
configuration.

Net yet complete, but integrated sufficiently to get the tests to pass.

And rename it to chain_initializer_interface

Aftermath of 55a84709 -- the chain
library no longer ever touches genesis_state_type; this type is only
recognized by native_contract now, so move its files from the chain
library to the native_contract library.

OK, so this turned out to be a **massive** refactor that I thought I'd
never finish, and no, I don't remember everything that changed here, but
I think it all worked out to be better than before, and almost all of it
was fairly necessary/the best of available options. Whew. I am SO glad
this is finally done.

High level goal: Continue refining the native contracts (ref #15),
specifically
walling off their state into their own contract DBs rather than globally
shared (read: stuff that race conditions are made of) general blockchain
state.

The first real change I made in this commit, which precipitated the
avalanche of changes that followed before the project became consistent
again, was to move the account balance out of account_object (globally
shared) to an object known only to the Eos Contract, thus eliminating
the possibility of race conditions/nondeterministic behavior.

The trouble is, balances are set at genesis. Genesis is processed by
chain_controller. chain_controller cannot know about balances, since
those are an abstraction defined on top of the Eos contract, which is an
abstraction defined on top of chain_controller. So all of genesis had to
be reimagined, and quite a lot of architectural changes had to be made
in order to create a logically consistent solution.

Changes (probably incomplete):
 - Fix up notifications within the native contract to support
precondition validation as well as application

 - Add notify handlers for CreateAccount to Eos and Staked Balance
contracts

 - Move account's liquid balance of EOS from account_object to
BalanceObject

 - Replace {producer,account}_object::id_type with AccountName most
everywhere except block_header, which still contains a
producer_object::id_type (potentially in violation of protocol
standards, but I want to confirm that before fixing it)
   - Reason: The name is not significantly slower, as it's fixed length so
no heap allocs, and it simplifies the code in quite a few places by
allowing us to look up objects directly rather than indirectly by
looking up an ID to get an intermediate object to get a handle for the
object we really wanted

 - Replace native_system_contract_plugin with native_contract library
   - Reason: The plugin was getting in the way. The native system
contract C++ implementation is simply too fundamental unless/until we
have a scripted implementation that works until the native
implementation gets installed

 - Completely reimagine genesis initialization, taking it largely out of
the hands of chain_controller and putting it in the hands of
native_contract
   - Reason: chain_controller understands relatively little about
genesis. It understands global_property_object and producer_object, but
not BalanceObject or StakedBalanceObject, etc... It also doesn't
understand the native_contract, and things like installing the native
contract, setting up accounts/balances, etc. all need to be handled by
something... native_contract is the most logical place to put it.

Sorry for the enormous commit... alas, this was the first time I got it
all building again and passing tests in days.

Create all three native contracts at genesis, and split the various
messages among them appropriately.

- remove BTS P2P code which depended on fc threads which depended on
deprecated boost::context API

- remove fc submodule and making fc part of eos repository

This completes the refactor (to a buildable/tests passing state) from my
last commit.

Rename database class to chain_controller, as this class manages
blockchain state and operations. Database was a misnomer.

Next step: make chain_controller a free-standing class, rather than
having it inherit from chainbase::database. This should give us a better
architecture, especially for testing, going forward.

Add support for producers specifying their votes on chain configuration
values to CreateProducer and UpdateProducer.

Add more tests around chain configuration, including a failing test case
verifying that the blockchain itself handles configuration updates
correctly. Presently this test fails because the chain does not yet
attempt to keep track of rounds, thus it cannot update the configuration
at the beginning of the round.

The macros to make testcases less tedious to write had opaque
abbreviated names like UPPDCR (my own fault, I know haha). Replace those
names with obvious ones like Update_Producer.

Also, sweep all the backend/implementation detail macros into
macro_support.hpp where they are hidden out of the way and unlikely to
be accidentally used directly.

We have a number of types defined in eos::chain like transaction,
message, etc. which parallel autogenerated types in eos::types. They are
identical in storage, but the chain types have various helper methods
which the autogenerated types, of course, lack.

I've unified these by letting the generated classes in types:: store the
data, and having the chain:: types inherit these data stores and add
functionality to them via methods.

Add helper macros for common testing activities including making {keys,
authorities, accounts, producers}, transferring funds, and updating
producers. Replace the boilerplate transaction jockeying in the tests
with these helpers. The tests are now much shorter and more readable. :)

The system contract C++ implementation is now in a plugin rather than
directly on database.

Note that testing_database installs this implementation on the testing
database, even though the test framework does not use appbase or appbase
plugins.

The simulated network was recursively calling push_block when
propagating blocks, which caused a lock timeout. Fix this issue by not
pushing a new block to the database that created it.

Also, replace the currently_propagating_block flag with boost's slot
blocker.

Bring in chain_plugin, net_plugin, http_plugin; wire appbase and plugins
into existing infrastructure, remove old app infrastructure.

Also remove transaction_evaluation_state and the size_checker program.

- Cleaned out some cruft
- Added some new macros to remove boilerplate
- Added some more test cases

Add a testing_network class to emulate a P2P network ferrying blocks
between databases.

Add some test cases exercising databases syncing with the
testing_network and resolving forks.