Obscure checkpoint/mempool interaction #2052

Currently, this is possible:

• Someone has a very old P2SH-invalid block in his block file
• When importing this block file, it gets accepted (because script checks that would catch this are disabled before the last checkpoint)
• It gets immediately reorganized when a later block is imported
• When disconnecting, its transactions enter the memory pool, as again, signature checks are disabled
• As long as not rebooted, this transaction remains in the mempool, and causes a (harmless) error to be reported when trying to construct a block

Two solutions (and I think both are useful anyway):

• During IBD, don't move disconnected transactions to the mempool at all
• Do not disable signature checks from disconnected transactions entering the mempool

Thanks to @gmaxwell for spotting a weird P2SH-invalid transaction in his mempool.

I like the belt-and-suspenders approach of doing BOTH dropping disconnected transactions during IBD and always checking signatures when adding disconnected transactions to the memory pool.

The second is especially important; I could imagine an edge case of a block chain re-org right at some rule-change threshold resulting in invalid-on-the-new-chain transactions in the memory pool.

Fixed by #2056