core: added checkpoint for last block

Changed the transaction pool to listen for ChainHeadEvent when resetting
the state instead of ChainEvent. It makes very little sense to burst
through transactions while we are catching up (e.g., have more than one
block to process)

* Add a checkpoint every X blocks
* Removed queued write

Add pendingTransactions and resend

core: reduce CPU load by reducing calls to checkQueue

* Reduced maxQueue count
* Added proper deletion past maxQueue limit
* Added cheap stats method to txpool

queueCheck was called for **every** transaction instead of:
1. add all txs
2. check queue

previously

1. add txs[i]
2. check queue
3. if i < len(txs) goto 1.

This fixes an issue with the lru cache not being available when calling
WriteBlock. WriteBlock previously always assumed to be called from the
InsertChain where the lru cache was always created prior to calling
WriteBlock. When being called from the worker this could lead in to a
nil pointer exception being thrown and causing database corruption.

Rebased peter's PR

core: optimisations

Conflicts:
	core/chain_manager.go
	miner/worker.go