Replay protection via CHECKSIG

Hi,

I thought of a possibly interesting way to prevent transaction replay in the event of a chain split, that seems better to the other approaches I've seen. Basically, update OP_CHECKSIG (and MULTISIG and the VERIFY variants, presumably via segwit versioning or using a NOP opcode) so that signatures can optionally specify an additional integer block-height. If this is provided, the message hash is combined with the block hash at the given height, before the signature is created/verified, and therefore the signature becomes invalid if used on a chain that does not have that particular block in its history [0].

It adds four bytes to a signature that uses the feature [1], along with a block hash lookup, and some extra sha ops when verifying the signature, but it otherwise seems pretty lightweight, and scales to an arbitrary number of forks including a pretty fair range of hard forks, as far as I can see, without requiring any coordination between any of the chains. So I think it's superior to what Johnson Lau proposed in January [2] or BIP 115 from last year [3].

Thoughts? Has this been proposed before and found wanting already?

Cheers, aj

[0] For consistency, you could use the genesis block hash if the signature
    doesn't specify a block height, which would lock a given signature to
    "bitcoin" or "testnet" or "litecoin", which might be beneficial.

[1] Conceivably a little less if you allow "-5" to mean "5 blocks ago"
    and miners replace a four byte absolute reference ("473000") with a
    one or two byte relative reference ("-206") when grabbing transactions
    from the mempool to put in the block template.

[2] https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2017-January/013473.html

[3] https://github.com/bitcoin/bips/blob/master/bip-0115.mediawiki